0% found this document useful (0 votes)

143 views18 pages

A Review On Machine Learning For EEG Signal

Uploaded by

huynhxuanthinh1911

We take content rights seriously. If you suspect this is your content, claim it here.

Available Formats

Download as PDF, TXT or read online on Scribd

0% found this document useful (0 votes)

143 views18 pages

A Review On Machine Learning For EEG Signal

Uploaded by

huynhxuanthinh1911

We take content rights seriously. If you suspect this is your content, claim it here.

Available Formats

Download as PDF, TXT or read online on Scribd

You are on page 1/ 18

This article has been accepted for publication in a future issue of this journal, but has not been

fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/RBME.2020.2969915, IEEE Reviews
in Biomedical Engineering

A Review on Machine Learning for EEG Signal

Processing in Bioengineering
Mohammad-Parsa Hosseini Senior Member, IEEE, Amin Hosseini Member, IEEE, Kiarash Ahi Member, IEEE

Abstract—Electroencephalography (EEG) has been a staple Clinically, EEG signals are used primarily to diagnose
method for identifying certain health conditions in patients since and treat various brain disorders such as epilepsy, tremor,
its discovery. Due to the many different types of classifiers concussions, strokes, and sleep disorders. More recent appli-
available to use, the analysis methods are also equally numerous.
In this review, we will be examining specifically machine learning cations of EEG include using machine learning as a method
methods that have been developed for EEG analysis with bio- of analysis. In particular, there is much research on epileptic
engineering applications. We reviewed literature from 1988 to seizure detection and sleep disorder research in combination
2018 to capture previous and current classification methods for with machine learning. Additionally, there is also a growing
EEG in multiple applications. From this information, we are able interest in studying EEG signals for gaming to control and
to determine the overall effectiveness of each machine learning
method as well as the key characteristics. We have found that all manipulate objects using brainwaves due to EEG monitoring
the primary methods used in machine learning have been applied for brain activity during tasks [40].
in some form in EEG classification. This ranges from Naive-Bayes EEG signals were first discovered in 1875 by Richard
to Decision Tree/Random Forest, to Support Vector Machine Caton, a physician who was studying electrical brain activity
(SVM). Supervised learning methods are on average of higher in rabbits and monkeys. Later on in the 1900s, the first human
accuracy than their unsupervised counterparts. This includes
SVM and KNN. While each of the methods individually is limited EEG recordings were made and studied with a focus on absent
in their accuracy in their respective applications, there is hope seizures. In the 1930s, epileptic spikes and seizure patterns
that the combination of methods when implemented properly has became noticed, which sparked EEG as a new field of interest.
a higher overall classification accuracy. This paper provides a EEG waveforms vary based on the band, which denotes
comprehensive overview of Machine Learning applications used the frequency range. The delta band is the slowest wave with
in EEG analysis. It also gives an overview of each of the methods
and general applications that each is best suited to. the highest amplitude, having a frequency range below 4Hz.
For adults, it is located frontally, while for children it is
Index Terms—Machine Learning, EEG, Survey, Medical Ap- located posteriorly. The theta band is between 4 to 7Hz and is
plications, Signal Processing, Signal Analysis.
most common in young children while signifying drowsiness
or arousal in adults. This band tends to spike due to an
I. I NTRODUCTION active inhibition of a movement or response. The alpha band
is between 8 to 14Hz, and it is correlated to eye muscle
Electroencephalography (EEG) is a method of testing elec- movements. It is located on both sides of the head’s posterior
trical signals in the brain. It is often applied as a technique for regions. The beta band is above 14Hz and is correlated with
data analysis such as time and frequency series analysis. The general motor behavior. It is located on both sides of the head’s
brain’s neurons contain ionic current, which creates voltage frontal regions [48].
fluctuations that EEG can measure. This electrical activity is Some of the advantages of using EEG compared to other
spontaneous and recorded over a period of time from many techniques to study brain function are low costs, tolerance to
scalp electrodes to form an EEG signal. [24] Traditionally, motion from subjects, and no radiation exposure risks. Some of
EEG signals are taken on the surface of the scalp, but there the disadvantages of using EEG include low spatial resolution
also exists iEEG signals, which are taken inside the brain. In and poor signal-to-noise ratio.
this paper, we will be focusing primarily on conventional scalp
EEG signals.
II. M ACHINE L EARNING M ETHODS FOR EEG
Conventionally, EEG recordings may be obtained by con-
necting electrodes to the scalp with the use of a conductive A. Overview
gel. A differential amplifier is then used to amplify each active Machine learning is the use of a set of mathematical models
electrode compared to the reference before it is sent through and algorithms to gradually improve the performance of a
an anti-aliasing filter. Finally, this filtered signal is converted singular task. It takes training data sets as input to use
with an analog-to-digital converter. as a guide for making estimates without being specifically
programmed to. The tasks vary widely in this space and
1 M.P. Hosseini is with Bioengineering Department, Santa Clara
can be categorized into two main groups: supervised and
University, CA, USA and AI Research, Silicon Valley, CA, USA
[email protected], [email protected] unsupervised learning. Unsupervised learning is the case when
3 A. Hosseini is with Electrical and Computer Engineering the algorithm builds a pattern of recognition from a data set
Department, Azad University, Central Tehran Branch, Tehran, Iran containing only inputs with no set outputs. Supervised learning
[email protected]
2 K. Ahi is An alumnus of the University of Connecticut, CT, USA has a subsection being semi-supervised learning. They are
[email protected] identical in the sense that they both learn from data sets

1937-3333 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: Northwestern University. Downloaded on May 03,2020 at 19:09:51 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/RBME.2020.2969915, IEEE Reviews
in Biomedical Engineering
2

between variables. Various types of regression include: uni-

variate and multivariate for quantitative response variables;
simple and multiple for predictor variables; linear for linearly
transformable data; nonlinear for nonlinearly transformable
data; analysis of variance for qualitative variable predictors;
analysis of covariance for the combination of qualitative and
quantitative variable predictors; and logistic for qualitative
response variable [90].
Legendre and Gauss first applied regression using the
Method of Least Squares. This method makes approximations
by summing the squares of each equation residual to best fit
the data, and it is applied in Linear Regression. as shown in
the equation below.

i = B0 + B1 xi + ei,i=1,...,n (1)
Linear Regression is one of the most common regression
techniques. In this model, the parameters are specified in
Fig. 1. Electrode placements in 10–20 system for EEG acquisition which the form of a linear combination, while each independent
describe the location of scalp electrodes in an EEG test.
variable is not necessarily linear. Multiple linear regression
is similar, except that there are several independent variables
with given inputs and known outputs with the exception that rather than just one. When the parameters are not linear,
semi-supervised has parts of the data set missing. Supervised nonlinear regression must be used. This also uses a sum of
learning is primarily used in applications of classification squares technique, though it uses an iterative procedure to
and regression while unsupervised learning lends itself to minimize the function.
feature learning and the inverse, dimensionality reduction. This
paper will discuss some of the most popular machine learning C. SVM
methods and categorize them based on the type of learning SVM is a subcategory of supervised learning used for
with some practical applications in EEG. analyzing data for classification and regression analysis. The
EEG signals can be used as indicators of harder to detect purpose is to map points in space such that the examples of
medical conditions with the assistance of machine learning the target categories are divided by the largest possible margin.
methods. In Fig. 2 the applications of machine learning on This allows for SVM to have a general lower generalization
EEG signals are shown based on supervised and unsupervised error as a classifier [43]. The objective is to find a hyperplane
learning. Supervised learning develops a predictive model or set of hyperplanes in an N-dimensional space. Support
using both input and desired output data is categorized to vectors are data points that are closer to a given hyperplane.
classification and regression which produce discrete and con- They maximize the margin of the classifier by changing the
tinuous accordingly. Unsupervised learning develops a predic- position and orientation of the hyperplane. Additionally, within
tive model using just input data is categorized to clustering this space, it is also possible that the points are not separable
and dimensionality reduction which produce discrete and linearly due to the position of the data. SVM is capable of
continuous accordingly. utilizing generated kernel functions or more commonly known
Fig. 3 describes the general flow of how machine learning as "kernel trick" to the data set to remedy this issue. This trick
is implemented to get the desired classification of the data involves the transformation of the existing algorithm from a
sets. The first step is signal acquisition. This is essentially lower-dimensional data set to a higher one. The amount of
the raw data, unedited. Pre-processing involves the removal information remains the same, but in this higher dimensional
of noise and other outliers in the data set. Feature extraction space, it is possible to create a linear classifier. Several K
determines the spectrum of the data point groupings and what kernels are assigned to each point which then help determine
features they correspond to. Feature selection is the isolation the best fit hyperplane for the newly transformed feature
of the desired classifiers that the machine learning method space. With enough K functions, it is possible to get precise
will be testing for the following training. Machine learning separation. The only major concern is overfitting. [120]. Fig. 4
training involves the use of training data sets, whether with depicts a sample of data separation in both 2D and 3D.
or without known outputs to refine the classification method.
Lastly, the testing phase is the processing of true test data sets →
−
w ·→
−
x − b = 1, −1 (2)
and comparing the overall accuracy of the desired feature.
Linear SVM classifier with hard margin

B. Regression Pl Pl Pl
1
W (α) = − i=1 αi + 2 i=1 j=1 yi yj αi αi xi xj (3)
Regression modeling is a popular tool in statistics be-
cause it is a simple way to create a functional relationship Kernel trick equation minimizing W subject to:

Fig. 2. Machine learning applications on EEG have been developed based on supervised and unsupervised learning in the literature. Supervised learning
is categorized to classification and regression which produce discrete and continuous accordingly. Unsupervised learning is categorized to clustering and
dimensionality reduction which produce discrete and continuous accordingly.

Fig. 3. The overall steps for EEG analysis by machine learning include preprocessing, feature extraction, feature selection, model training, model testing.

TABLE I
R EGRESSION MODELS APPLIED FOR EEG ANALYSIS

Author(s) Machine Learning Method Application Data Set Results

Rajaguru et al., 2017 [95] Logistic Regression Epilepsy Classification 20 patients Performance Index 91.39%
and Accuracy 95.88%
Kim et al., 2014 [68] Non-Linear Regression Reconstruction of hand 4 subjects not listed
movements from EEG
signals
Dora et al., 2016 [27] Linear Regression Robust ECG Artifact Re- not listed Accuracy 98.11%
moval
Rajaguru et al., 2017 [94] LRGMM Epilepsy Classification not listed Accuracy 97.91%
Murakami et al., 2015 [82] Logistic Regression Motion Discrimination 3 subjects Single Threshold Process-
ing Accuracy 77.0%
Li et al., 2015 [75] Logistic Regression Ocular Artefacts Correction 68 subjects not listed
Method for Discriminative
EEG Analysis
Dong et al., 2013 [26] Linear Regression Visual Attention Modeling 6 subjects not listed
Jain et al., 2016 [60] Auto-Regression Fatigue Detection and Esti- 14 subjects not listed
mation
Hu et al., 2016 [56] Auto-Regression EEG Authentication System not listed Accuracy 92.93%
Hamilton et al., 2015 [35] EBMAL Regression Offline EEG-Based Driver 16 subjects not listed
Drowsiness Estimation
Struck et al., 2017 [108] Logistic Regression Seizure probability in hos- 4772 Participants area under the curve of
pitalized patients 0.819 and average calibra-
tion error of 2.7% (95% CI,
2.0%-3.6%)
Roy et al., 2018 [98] Logistic Regression, neural networks, CNN, RNN Automatic Abnormal EEG 1488 abnormal, 1529 normal EEG Deep gated RNN achieve
Identification 3.47% better performance
than previously reported re-
sults

Pl Compared to other supervised and unsupervised learning

i=1 yi αi = 0 techniques, neural networks deliver the most optimum results,
(4)
0≤αi ≤C by segregating data to the deepest level intended. The modern
generation has been successful in identifying and documenting
D. KNN K-Nearset Neighbours many techniques involving the use of Artificial neural net-
KNN is one of the supervised machine learning algorithms. works for analyzing and classifying the EEG signals and sim-
In supervised learning, the relationship between the input and ulate the results for serving different medical conditions [85].
output is already established for the training data set, i.e for a EEG signals or commonly called as Electroencephalogram
given input the output is already known. Supervised learning are set of highly complex signals which are studied to find
is categorized into regression and classification. KNN can abnormalities in a human brain activity [101]. The latter is
be used for both classification and regression. The input for measured when neurons in a brain start firing which produces
both classification and regression is the same but the output an electric current within the dendrites. This electric current,
differs respectively. Example input-output pairs are used for therefore, produces a magnetic field all over the scalp which
predicting the output for untrained data set. KNN classifies the is recorded through a signal generator that gives the activity at
input based on the classification of its K neighbors. To find the any given location in a brain. These signals need to be studied
nearest neighbors, Euclidean distance or Mahalanobis distance and segregated for the type of abnormality we are focusing
is calculated from the input to all known data points. After the on, and for that, we are discussing different Machine learning
distance is calculated, K nearest neighbors are selected. It then models that can be used for different EEG applications. The
classifies the input based on similarities between input and its need for proper classification and analysis gets itself the prior-
K- neighbors. The selection of K is based on the size of the ity when we compare the traditional method of EEG analysis.
data set. The square root of the size of the data set is taken and The standard procedure involves time-frequency analysis and
if the result is an even number then 1 is added or subtracted spectrogram image processing for analyzing EEG signals. The
from it. The result is then established as K for that data set. K characteristic waveform of EEG fall on specific frequency
is selected to be an odd number to avoid bias in the prediction bands namely: alpha (8 to 15Hz), beta (14 to 30HZ), theta (4 to
of input. 8Hz) and delta (< 4Hz), and previous classification methods
like FFT (Fast Fourier transforms) have been characterized
E. ANN to have very high noise sensitivity which limits the ability
to analyze the signals effectively. This brings the spotlight to
Neural networks, commonly called the Artificial neural
implement Neural networks or ANN for measuring the activity
networks in the computing world, is a mathematical model
of EEGs effectively. The current section discusses how ANN
very similar to the structure of neural networks seen in a
can be applied for the synthesis of EEG data, and classify
human brain. To understand how the model works, researches
different signals based on their attributes and features. As we
have put forth several theories and examples showing the
know by now, ANN works by analyzing the data set provided
interaction between different layers of the neural networks to
against multiple possibilities and connections to deliver an
convert the given input into the desired output [5].
optimum output as required [31]. The main characteristics of
Imagine you are at a bar, and looking at the menu to order a
the network architecture depend on a few highlighted factors
nice beer. Your favorite is IPA and as soon as you see that on
show below:
the list, you order it. So what happened in your brain is that
1. Type of input signal (signal dimensionality and behavior)
you provided multiple inputs for beer choice to your brain’s
2. Topology of connections
neural network, IPA choice had a preferable weight as that
3. Interactions between different network layers
being your most favorite beer; the brain made a decision and
4. Mode of operation
gave you the output. This is a basic example of how neural
5. Output interpretation
networks operate. The architecture of the model shows the
As each application varies and has to have a specific ap-
decision-making process which involves much deeper layers
proach – Long term or short term EEG segment analysis, real-
of interaction that lie between the input and the output layer.
time process or time-delayed process, type of EEG channel
Fig. 5 shows the classification of different layers, of ANN.
analysis (single or multiple) – which can be easily targeted and
From the Mathematical model of neural networks, it is
synthesized using ANN. Once the EEG signals are converted
evident that any given input at the first layer undergoes a
to waveforms in user-friendly GUIs, the classification of these
function put forth by the algorithm which will narrow down
signals happens with ANN, with the selection of a particular
multiple combinations and options, to depict the desired out-
type of network for a specific use case – Feedforward back-
put. This can be observed in Fig. 6. For ANN, the classification
propagation, Radial basis function, Recurrent Neural networks.
technique can be brought about by:
It is important to see how different types of ANN operate and
Summation of input-weight product and Bias
Pn the architecture which facilitates that operation.
i=1 (wi xi ) + bias (5) 1. Feedforward Neural Networks: This is a type of network
Activation Layer where data flows in only one direction, starting from the
( P input nodes, passing through the hidden nodes and arriving
1if wx + b ≥ 0 at the output nodes. This network ensures no loop or cycle
Output = f (x) = P (6)
0if wx + b < 0 formation, making the information flow in a specific direction

Fig. 4. Higher dimension kernel separation. The kernel trick involves the transformation of the existing algorithm from a lower dimensional data set to a
higher one.

TABLE II
S UPPORT V ECTOR M ACHINE A PPLICATIONS WITH EEG

Author(s) Machine Learning Method Application Data Set Results

Jalilifard et al., 2016 [64] SVM Emotion Classification 19 patients Accuracy 96.83%
Sai et al., 2018 [100] SVM EEG artifact removal 11 patients Accuracy 99.1%
Zhang et al., 2017 [122] SVM Seizure Detection of EEG Not Reported Accuracy 98.1%
Torabi et al., 2017 [111] SVM, KNN Multiple Sclerosis Detection Not Reported Accuracy 93.08%
Aghajani et al., 2017 [3] SVM EEG Measuring Mental Workload 17 patients Accuracy 90.0%
Amin et al., 2017 [9] SVM EEG Classification 48 practice patterns Accuracy 98.57%
Jaiswal et al., 2017 [62] SVM Epilepsy Detection of EEG 7 patients Accuracy 100%
Huang et al., 2017 [57] SVM Multimodal Facial Recognition 40 patients Accuracy 82.75%
Ahani et al., 2014 [4] SVM Meditation EEG Detection 34 patients Accuracy 85%
Mumtaz et al., 2018 [81] SVM Alcohol Use Disorder Detection 30 patients and 30 normal Accuracy 98.0%
Dian et al., 2015 [25] SVM Identification of brain regions of in- 6 patients Accuracy Proposed method
terest for epilepsy surgery planning is scalable across multi-
ple patients exhibiting En-
gel Class I outcomes
Zhuang et al., 2018 [123] SVM Emotion recognition 30 participants research lays a substan-
tial foundation for real-time
recognition of comprehen-
sive endogenous emotion%
Beganovic et al., 2018 [13] SVM, KNN epileptic seizure occurrence 20 patients, dimensionality reduction/selection %

only. Fig. II-E shows the architecture for Feedforward network For a multiquadric RBF with a scalar input can be shown
mechanism as:
2. Radial basis function: In the field of artificial neural √2
r +(x−c)2 (8)
networks and mathematical modeling, RBF is a type of ANN h(x) = r
which makes use of radial basis functions (An arbitrary real-
In this case the Gaussian RBF increases with increase in
valued function, the value of which is determined by functions
the distance from the center.
location from the origin). Thus, the network determines the
3. Recurrent Neural Networks: As the name suggests, RNN
output by a linear combination of RBF of the inputs and
is a type of Artificial Neural Network which has connections
parameters given for the neurons. As shown in Fig. 8 the
between different nodes, with a specific assigned direction for
structure operates by summing the centers/widths of the points
output flow to a specific node. Here, the flow of data can form
with the associated weights to get us the final output.
loops and cycles to feed the data back to a specific node as
A typical RBF is a Gaussian distribution, in case of a scalar
intended. This technique is illustrated in Fig. 9 which shows
input, and is given by:
the backpropagation of information from one layer to another
and to a specifically intended node.
2
h(x) = exp( −(x−c)
r2 ) (7) To understand the working of RNN it is important to define
the transitions from one previous state to a new state. Let
Where c is the center, and r is the radius parameters. A Xt be the input vector, Ht be the new state, and Ht-1 be the
Gaussian RBF distribution decreases as the distance from the previous state. RNN is observed to be a function of the input
center increases. vector and the previous state, which will land us to the new

Fig. 7. Feedforward Neural Network. There are two directions for information
Fig. 5. A feedforward neural network with input layer, hidden layers, and flows, forward propagation and backpropagation. Forward propagation is used
output layer. in the prediction time while backpropagation is used for adjusting the weights
to minimize loss.

Fig. 8. A radial basis function network is an ANN which uses radial

basis functions as activation functions. A linear combination of radial basis
Fig. 6. Structure breakdown for ANN with three types of nodes, input nodes, functions of the inputs and the parameters of neurons is used for the output
hidden nodes, and output nodes. of the network. These structures have many applications such as time series
prediction, classification, and function approximation.

state Ht. We can represent a simple Vanilla version of the

RNN by obtaining the weight function Fw and implementing F. Naive Bayes
that to find the output function Yt. This can be represented as
follows: Naive Bayes classifier is a popular text categorization
method that applies Bayes’ theorem to separate data based on
ht = fw (ht−1 , xt ) (9) simple trained features. Essentially, the model assigns labels as
feature vectors within a finite set. While simple in nature, with
adequate pre-processing it can match more advanced methods
ht = tanh(Whh .ht−1 + Wxh .xt ) (10) such as SVM discussed above. The one disadvantage of the
By applying the tan hyperbolic function the dot product of naive Bayes method is that it considers all of the feature
associated weights from previous states and the dot product vectors as independent from one another regardless of any real
of associated weights and input state, we shall have the value correlation. The main advantage of it is that it only needs a
of the new state. We can have the final output function as: small number of training data sets to begin correctly estimating
the parameters necessary for classification. Several models can
yt = Why .ht (11) be implemented for the Bayes method. The most common of
which is the probabilistic model. In this model, the features are

TABLE III
A RTIFICIAL N EURAL N ETWORKS APPLICATION FOR EEG A NALYSIS

Author(s) Machine Learning Method Application Data Set Results

Hramov, Alexander E., et al., 2017 [53] ANN Perceptual Interpretations of a Not Reported Accuracy 95%
Bistable Image
Tzallas, A. T., Tsipouras, M. G. et al. 2007 [114] ANN Automatic seizure detection . Not Reported Accuracy 97.72%
Sharma, A.,Tewari R. P., et al. 2018 [103] ANN Epileptic seizure anticipation Not Reported Accuracy 92.3%
Saini, J. and Dutta, M., et al. 2018 [101] ANN Epilepsy classification using opti- 100 Samples Accuracy 99.3%
mized artificial neural network.
Chiarelli, Antonio Maria, et al. 2018 [20] ANN Deep learning for hybrid EEG. 15 Participants Accuracy 99.3%
Lee, Y. H., Hsieh, Y. J., Shiah, Y. J., et al. 2017 [73] ANN, SVM EEG Cross section Analysis 10 Samples Accuracy 98%
Guo, L., Rivero, D., Pazos, A., et al. 2010 [33] ANN Epileptic seizure detection Public Not Reported
Ogulata, S. N., Şahin, C., Erol, R., et al. 2009 [85] ANN Classification of primary general- 4 groups Accuracy 78-98%
ized epilepsy by EEG signals..
Srinivasan, V., Eswaran, C., Sriraam, N., et al. 2007 [107] ANN Approximate entropy-based epilep- Public Accuracy 100%
tic EEG detection.
Ghosh-Dastidar, S., Adeli, H., et al. 2009 [31] ANN Multiple spiking neural networks Not Reported Accuracy 90-94%
for application in epilepsy and
seizure detection.

Fig. 9. Recurrent Neural Network where connections between nodes form Fig. 10. Example for decision tree technique to determine a health condition
a directed graph along a temporal sequence. It makes previous outputs to be
used as inputs.
G. Decision Tree and Random Forest
represented by vectors and it assigns probabilities to a given Decision trees use questions about the features of an item to
outcome or case. Event models can be separated into 2 main classify data. Each question can be represented as a node, in
classes, Gaussian Naive Bayes and Multinomial Naive Bayes. which there is a child node for each answer to that question.
In a data set with continuous values, a good assumption would This creates a hierarchy, in other words, a tree. The most basic
be that it follows a Gaussian distribution. Using this method tree would be a binary one in which each question results in a
the Bayes method assigns probabilities based on the curve. A yes or no answer. Therefore there is a yes and no child node
multinomial event model represents the frequencies of specific for each parent node question. Data is sorted through the tree
events spawned from multinomials, often as a histogram. A by starting at the top-most node, also known as the root, and
potential concern is when a feature does not occur in the data maneuvering its way down to the leaf, or the node that has no
set at all. This causes the multiple of all the estimates to be children. The path taken is dependent on the data’s features.
zero. It can be corrected with a pseudocount to smooth out Once the data reaches the leaf, it can be classified under the
any outliers in the data set [97]. class associated with that particular leaf [69].
The advantages of decision trees are that they are simplistic
P (x)|(c) and can be easily combined with other techniques for decision
P (c|x) = P (x) (12)
making. The disadvantages of decision trees are that they
The probabilistic Naive Bayes Model are somewhat unstable as well as inaccurate, especially with
(v−µk )2
varying level sizes which cause biases towards larger levels.
−
2σ 2
P (x = v | Ck ) = √ 1 2
e k (13) In the study of machine learning, and different classifying
2πσk
and distribution methods, we come across the Random Forest
The Gaussian Naive Bayes Model technique, which can be used for both data classification and

TABLE IV
NAIVE BAYES A PPLICATIONS WITH EEG

Author(s) Machine Learning Method Application Data Set Results

Amin et al., 2017 [9] Naive Bayes EEG Classification 48 practice patterns Accuracy 81.07-91.60%
Fan et al., 2015 [29] Naive Bayes EEG autism detection 16 patients Accuracy 65-76%
Rytkönen et al., 2011 [99] Naive Bayes Sleep Scoring 2 humans and 30 animals Accuracy 92%
Biswal et al., 2015 [16] Naive Bayes Automated Information Extraction 42,972 reports Accuracy 97.53%
Mumtaz et al., 2018 [80] Naive Bayes Major Depressive Disorder 30 patients and 30 normal Accuracy 93.6%
Fallani et al., 2011 [28] Naive Bayes Subject Recognition 50 subjects Accuracy 78-89%
Laton et al., 2014 [71] Naive Bayes Schizophrenia Subject Recognition 54 patients and 50 normal Accuracy 79.8%
Bigdely et al., 2008 [15] Naive Bayes Brain Activity Classification 7 subjects Accuracy 87%
Sharmila et al., 2017 [105] Naive Bayes Epilepsy Detection Not Reported Accuracy 98.6%
Biswal et al., 2015 [16] Naive Bayes Automated information extraction 3277 documents Accuracy The average [95%
from free-text EEG reports CI] area under the receiver
operating curve was 99.05
[98.79, 99.32]% for detect-
ing reports with seizures,
and 96.15 [92.31, 100.00]%
for detecting reports with
epileptiform discharges
Gao et al., 2016 [30] KNN, naive Bayes, SVM automatic sleep scoring in mice 16 mice At 1% rejection rate, the
algorithm matches the accu-
racy of a human scorer
Sharmila et al., 2018 [104] Naive Bayes, SVM detection of epileptic seizure Not Reported Accuracy 100%
Page et al., 2015 [88] Naive Bayes, SVM, KNN ultra-low power feature extraction 10 Patients Accuracy 100%
and classification system for wear-
able seizure detection
Combrisson et al., 2015 [21] Naive Bayes, SVM The caveat of theoretical chance Not Reported Accuracy 70%
levels in brain signal classification
and statistical assessment of decod-
ing accuracy
Jirayucharoensak et al., 2014 [65] Naive Bayes, SVM, Deep Learning Network emotion recognition 32 Subjects Accuracy 49.52%

regression operations. As the name suggests, Random Forest different sized classification trees. This highly depends on the
operates by producing a multitude of decision trees and trained number of random variables which will be randomly selected
by performing bagging operation to combine multiple decision to form the classification tree. One of the many trees describes
trees or models to arrive at a more stable and accurate data a series of questions pertaining to the condition under study,
prediction. Random Forest creates additional randomness to where the next set of questions depend on the answer from
the data being structured; i.e. instead of finding the most the query. The solution from all the trees is summed up and
important feature from the given set, it operates to find the best classification happens through a majority voting where the
feature among a random set of a defined subset of features. best suitable classification is chosen. However, if the trees are
This results in a more diverse and better result model. found to be unstable, where minor changes in the data set can
change the whole decision tree, we might end up with a wrong
One of the most recognized and well-known application for classification. Combining the RF model with other Machine
EEG analysis is for diagnosing epileptic seizures. Usually, Learning models have shown an improvement in analyzing
the diagnosis is carried out during an epileptic attack, however, EEG signals more precisely and accurately.
the current process involves doctors who diagnose the problem
during a post-attack period which makes it a difficult process
and often leads to wrong conclusions. This created an opening
for analyzing the condition through the generation of EEG H. Ensemble Learning
signals, by recording the activity of the brain, by measuring the
discharged electric signals from the neurons, and calibrating Ensemble learning is a supervised learning algorithm. As the
those results in a waveform. Though the method may seem name suggests, ensemble learning ensemble’s many different
practical and computational, this is known to be one of the algorithms to make a model that gives a better predictive
most sophisticated processes to arrive at accurate results. This performance. The general idea is to improve the overall
has led to the development and implementation of Machine performance by combining decisions received from different
Learning models to analyze the EEG signals and classify them multiple models. It is based on the concept of diversity, more
according to the most appropriate medical condition under diverse models are considered for obtaining the results for
study. Earlier we discussed how different models contribute the same problem in comparison to single models. This gives
towards this cause, and how those models interact with the a set of hypotheses which can be combined to gain better
given set of signals. But when it comes to applying the performance. All the single models are called as base learners
Random Forest model, the process is different, where the when combined are called as an ensemble. The ensemble is
classification happens by a majority voting method, for the mostly better than the base learners from which the ensemble
decision trees which have been formed. The process of ran- is made. Ensemble learning can be used in the fields of
domization takes in not only the sample data provided, but medicine, fraud detection, banking, malware and intrusion
also on the predictor variables, which leads to a collection of detection, face and emotion recognition, etc.

TABLE V
R EVIEW ON DECISION TREE AND RANDOM FOREST

Author(s) Machine Learning Method Application Data Set Results

Rajaguru et al., 2017 [96] SDT Epilepsy Classification not listed Accuracy 96.83%
Ishfaque et al., 2013 [58] DT Brain Computer Interface not listed Accuracy 81.6%
Jakaite et al., 2010 [63] DT Newborn Brain Maturity 200 patients and 100 normal Accuracy 86.5%
Anastasiadou et al., 2017 [10] Random Forest Scalp Recordings for Automatic not listed not listed
Muscle Artifact Detection and Re-
moval
Hu et al., 2018 [55] GBDT Automated Driver Fatigue Detec- 22 subjects Accuracy 94.0%
tion
Vijayakumar et al., 2017 [116] Random Forest Quantify and Characterize Tonic not listed Accuracy 89.45%
Thermal Pain
Le et al., 2017 [72] Random Forest Surface and intracranial EEG spike 17 scalp patients and 10 intracranial Accuracy 62% recall and
detection 26% precision for surface
EEG subjects and 63% re-
call and 53% precision for
intracranial EEG subjects
Bentlemsan et al., 2014 [14] Random Forest Motor Imagery Classification 9 subjects Kappa 0.59
Wang et al., 2013 [117] Random Forest Classification of Neonatal 209 normal and 73 abnormal infants Accuracy 91.46%
Amplitude-Integrated EEG
Weichwald et al., 2014 [118] Random Forest Decoding Index Finger Position 12 subjects Accuracy 12.29%
from EEG
Hamilton et al., 2015 [35] Boosted Rotational Forests Eye State Prediction not listed Accuracy 97.4%
Bose et al., 2017 [18] Random Forests Seizure detection 5 patients Accuracy 98%

TABLE VI
A REVIEW ON ENSEMBLE LEARNING STATE OF ARTS

Author(s) Machine Learning Method Application Data Set Results

Prabhakar et al.,2015 [89] KNN Diagnosis of multiple scle- Not reported Accuracy For the direction-based and the
rosis detection color-luminance-based tasks, maximum clas-
sification performances were 93.08 and
79.79% respectively
Gunay et al., 2018 [32] KNN, Naïve Bayes Epilepsy Disorder 500 Accuracy 73% for KNN and 92% for Naive
Bayes
ozerdem et al., 2017 [87] KNN,SVM Emotion recognition 32 healthy subjects 72.92%
Sharmila et al.,2017 [105] PCA, LDA with K-NN Wavelet-based feature ex- Not reported Accuracy PCA, LDA with K-NN achieves
traction for classification of 98.5% and 100%
epileptic seizure EEG
Manjusha et al.,2016 [77] K-means, KNN Performance analysis of 20 Patients Accuracy A high Quality value of 22.37 with
KNN classifier and K- K-means clustering and a low value of 18.02
means clustering for robust are obtained with KNN classifier
classification of epilepsy
Tuyisenge et al.,2018 [113] ensemble learning Automatic bad channel de- 206 patients Accuracy 99.77% for 110 patients
tection
Hosseini et al., 2018 [48] ensemble learning EEG classification 8 patients Accuracy Using leave-one-out cross-
validation, the accuracy, sensitivity,
specificity, and both false positive and
false negative ratios of the proposed method
were found to be 0.97, 0.98, 0.96, 0.04, and
0.02, respectively.
Al Zoubi et al., 2018 [6] stack-ensemble learning Predicting Age From Brain 468 Healthy and 297 female pa- Accuracy The stack-ensemble age prediction
EEG Signals tients from Tulsa-1000 model achieved R2 = 0.37 (0.06), Mean Ab-
solute Error (MAE) = 6.87(0.69) and RMSE
= 8.46(0.59) in years
Antoniades et al., 2018 [11] ensemble learning Mapping Scalp Intracranial EEG (iEEG) data Accuracy classification accuracy of 68% an
increase of 6% over the previously proposed
linear regression mapping
Hassan et al., 2016 [37] ensemble learning Automatic identification of segments of EEG signals Accuracy Proposed seizure detection scheme
epileptic seizures performs better than the existing works in
terms of accuracy, sensitivity, specificity, and
Cohen’s Kappa coefficient
Meyer et al., 2014 [78] random forest ensemble classifier Predicting motor learning 6 patients Accuracy learned models successfully gener-
alized to novel subjects
Smart et al., 2015 [78] Smart et al., 2015 Semi-automated patient- 24 patients classification via centroid-based clustering
specific scalp EEG seizure methods, such as k-means and k-mediod algo-
detection rithms, or agglomerative clustering methods
appear best suited for scalp EEG seizure-
detection applications
Iturrate et al., 2010 [59] reinforcement learning Robot reinforcement learn- 5 classes it is possible to apply RL using EEG based
ing using EEG-based re- reward signals
ward signals

of truth for a given set of input variables. Fuzzy logic is

observed to be implemented in many applications in the
field of Machine learning, right from control theory to AI
(Artificial Intelligence). This is an exact simulation of a human
being’s brain for thinking and reasoning abilities [115]. For a
Fuzzy system to work effectively, the following features and
components need to be assured of performance:
1. Fuzzy sets: A fuzzy set is considered to be correspondent
with the member function, which is defined in a fuzzy space
where the variables are set. The feature of a member function
is to provide a degree of membership to any element within
the well defined fuzzy sets. Then the member function assigns
these elements a numerical value between 0 to 1, where 0
implies the corresponding element is not an element in the
fuzzy set or 1 means the corresponding element is an element
of the fuzzy set. One main aspect here is that the elements are
allowed to be assigned a partial membership, stating that the
corresponding element can be an element from the fuzzy set
to a certain degree ranging from 0 to 1. However, this criterion
is only satisfied according to the set of defined Fuzzy Rules.
Fig. 12 shows the breakdown of a typical fuzzy system
2. Fuzzy Rules: The way a fuzzy logic is intended to
function is defined by a set of applied fuzzy rules, which
determines the output which will be specified by the IF-
Fig. 11. Random Forest is an ensemble learning method which is used THEN rules. The IF-THEN rules are observed to create a
mostly for classification and regression. It operates by creating a multitude conditional statement that will consist of fuzzy logic. For
of decision trees on various sub-samples of the dataset and uses majority example, the IF-THEN assumes where X and Y are intended
voting or averaging for finding output. This model improves the accuracy of
prediction and can control over-fitting. terms and are evaluated by the terms of fuzzy sets with the
range being U and V. This divides the statement into two
parts namely antecedent and consequent. If the antecedent is a
I. Fuzzy Logic preceding statement which specifies the terms X and U, then
the consequent statement should conclude with Y and V. These
Fuzzy Logic is a machine learning technique that segregates combined makes a rule which states: if X is U, then Y is V.
the given data by assigning a range of truth values between However, these rules are based on the natural language and
0.0 to 1.0. This method does not merely accept or reject data model representation, based on the given fuzzy sets and logic.
by assigning a definite truth or false value (binary value 0 or 3. Fuzzy Logic Inference or Fuzzy Inference System (FIS):
1). Instead, it decides upon justification by assigning a degree Once the set of fuzzy rules and membership functions have
of truth value, where the data is not completely true, but to been defined, the FIS is implemented for process simulation,
an extent can be considered true by a certain value. Consider and control, and is done by the type of data or knowledge
the following example: We have two types of known shades provided. The FIS system usually operates on 3 stages: In the
of blue color which are light blue and dark blue. If you are in first stage, the numerical input variables which are provided to
a process of training your computer to determine which exact the system, are mapped for a degree of compatibility for the
shade of blue a given input (sea blue) might be, we can assign respective fuzzy sets. This is called the Fuzzification process.
a truth value for that input ranging from 0.0 to 1.0 which will This process allows the system to express the input and output
state that the input can be 80 percent closer to the intended in fuzzy-readable linguistic terms. In the second stage, the
true color. system processes the rules according to the strengths of each
Almost every household machine or equipment (like the air input variable. in the third stage, the resulting fuzzy values
conditioner, washing machine, etc.) operates on the concept are converted back to numerical values, by the process of
of Fuzzy Logic. This logic is fed to a control system usually Defuzzification. This process thereby maps the fuzzy domain
called the Fuzzy system control, where each component is output back to the crisp domain, which makes the output clear.
designed to function and alter another physical operating 4. Fuzzy Score: The output from the FIS system is in the
system, to achieve the desired functionality. To understand how form of a fuzzy score, for all the individual input scores that
a fuzzy system works, it is necessary to analyze the system are known to be generated by the system. The FIS system
requirements and the intent for using a fuzzy system [22]. To calculates the fuzzy score by taking into considerations all
make a system a knowledge-based functioning element with the defined fuzzy constraints and membership functions. The
the capacity to apply the human cognitive processes, such score is dependent on the type of rules applied and the type
as reasoning and thinking, has to have a stable component of input variables. Every input variable is assigned a score by
that can provide output on the perspective of the degree the FIS based on the fuzzy rules criteria. Typical fuzzy system

Fig. 12. Example for Fuzzy System. For a Fuzzy system to work effectively, the following features and components needs to be assured of performance: 1.
Fuzzy sets, 2. Fuzzy Rules, 3. Fuzzy Logic Inference, 4. Fuzzy Score.

Fig. 13. Operation of Fuzzy System. The input variables are mapped by fuzzy sets which are sets of membership functions. This process of mapping a crisp
input value to a fuzzy value is named fuzzification.

TABLE VII
F UZZY L OGIC FOR EEG A NALYSIS

Author(s) Machine Learning Method Application Data Set Results

Li Peng., et al., 2018 [74] Fuzzy Logic Detection of epileptic seizure Public Accuracy 93%
Abbasi Hamid., et al. 2016 [2] Fuzzy Logic Stereotypic evolving micro-scale Not Reported Accuracy 78.71%
seizures (SEMS) identification .
Rabbi A. F., Azinfar, L., Fazel-Rezai, R, et al. 2013 [91] Fuzzy Logic, ANN Seizure prediction using adaptive Not Reported Accuracy 80%
neuro-fuzzy inference system
Rabbi A. F., Fazel-Rezai, R., et al. 2012 [92] Fuzzy Logic Seizure onset detection in Intracra- 20 Patients Accuracy 95.8%
nial EEG.
Aarabi, A., Fazel-Rezai, R., Aghakhani, Y., et al. 2009 [1] Fuzzy Logic Seizure detection in Intracranial 21 Patients Accuracy 98.7%
EEG using a fuzzy inference system
Cosenza-Andraus, M. E., et al. 2006 [22] Fuzzy Logic Video-electroencephalography pro- 22 Adult patients Accuracy 91%
longed monitoring
Sharif, B., Jafari, A. H., et al. 2017 [102] Fuzzy Logic Prediction of seizures from EEG 19 Patients Sensitivity 91.8-96.6%
sugnals
Hsu, Wei-Yen. , et al. 2015 [54] Fuzzy Logic Assembling multi-Feature EEG Not reported Accuracy 88.2%
classifier
Ubeyli, E. D., et al. 2006 [115] Fuzzy Logic Fuzzy similarity index for discrim- 5 Patients Not Reported
ination of EEG signals.
Subasi, Abdulhamit., et al. 2007 [109] Fuzzy Logic, ANN Epileptic seizure detection using Public Not Reported
wavelet feature extraction

operation is illustrated in Fig. 13. density and regression estimation, pattern recognition, and data
analytics [?]
As the main application of Machine Learning is found to
be in pattern recognition of EEG signals, Fuzzy Logic can
be used to determine the correct recognition rate of EEG J. Linear Discriminant Analysis
classifications at different stages. However, a combination of For a given data set with a wide selection of random vari-
Fuzzy logic with Neural networks often called the Neuro- ables, it is necessary to perform a dimensionality reduction to
Fuzzy system, is adopted, where the system can apply the reduce the number of parameters to specific principle variables
fuzzy parameter (like fuzzy sets, fuzzy rules) and combine to reduce the dimensional space of the dataset. As there are
that with the neural network approximation techniques for many possible ways to classify the data, the dimensionality
extensive analysis. The Neuro-Fuzzy system [91] is found reduction technique is implemented by two techniques: The
to be highly beneficial for medical condition diagnostics, Principle component analysis, and linear discriminant analysis.

Both PCA and LDA have similar functionalities and appli- small. Some of the disadvantages of K-means include the high
cations. However, the LDA technique can handles situations significance of initial conditions on final outputs, sensitivity to
where the within-class frequencies need not be equal and the scaling, and a correlation between data order with final results.
standout factor is that it offers a high ratio and significant
separation between the between-class variance, and the within- L. Reinforcement learning
class variance. The main difference between the PCA and LDA
being, PCA is more applicable for classification of features, Machine learning can be divided into three categories: Su-
and LDA is applicable for data classification. pervised learning, Unsupervised learning and Reinforcement
The most common technique used for dimensionality reduc- learning. Reinforcement learning uses the given data to choose
tion is Linear discriminant analysis (LDA). The main criteria an action for an environment that yields the maximum ex-
behind this technique are to offer a good separability between pected long term reward. Reinforcement learning concentrates
different classes and to avoid overfitting of the curve. This more on performance than the given data. Concerning the
will significantly reduce computational costs and provides given data, Reinforcement learning can be described as a
better classification, by projecting the given feature space combination of supervised learning and unsupervised learn-
with n-dimensional samples onto a precise and smaller feature ing. Reinforcement learning may or may not have known
subspace. In a typical PCA analysis, the location, shape, and input/output pairs of the result. Reinforcement learning tries to
structure of the data set completely change. But for LDA, find a combination between exploration of unknown and ex-
the technique maintains the location and shape of the data set ploitation of current knowledge which also poses exploration
when transformed into a different smaller space. This happens vs exploitation trade-off. The biggest problem in modern-
through defining a set of vectors on the transformed space to day brain-computer interface (BCI) systems is that the per-
distinguish and separate. In an LDA technique this usually formance factor of these systems in controlling a BCI can
happens by two different approaches: and will decrease significantly over the period. Due to this
1. Class-independent transformation: This approach mainly issue, the necessity of controlling a BCI has increased, and
focuses on increasing the ratio of overall variance to the the motivation factor behind this is quite low. To eradicate
within-class variance and it only uses one criterion to optimize this scenario and find a solution to the addressed problem, we
the process of data set transformation. This transforms all the must enable a continuous feedback system from the subject
necessary data points irrespective of their class. So here, each and feed that to a Reinforcement learning agent to train and
class is observed to be separate from all other classes. support the case in finding an accurate solution. The purpose
2.Class-dependent transformation: Here, the main objective here is to use the RL agent to control the actions of the
is to increase the ratio between the class variables to that of the given task and as the process precedes, the supporting impact
within-class variables, to offer a sufficient range of separability from the agent is decreased and the subject will take over
for classification the control mechanism. As the subject takes over the control,
For the application of analysis of EEG signals and the Brain- the criteria are to maintain the subject at the state and to
computing interface, the exploration of advanced methods to measure the performance by implementing a reward system
separate and segregate the data sets with multiple variables, in that assigns certain points to the subject on how well it controls
an effective manner. A received EEG signal may be distorted the task without any agent present. The main objective of the
by noise disturbance and may have to be separated effectively, reinforcement agent is to interact with the subject in uncertain
to achieve accurate results. For this purpose, the technique of conditions, and maximize the numerical long term reward for
dimensionality reduction is being implemented, to reduce the the subject, basically taking a subject from one state to another.
data set and separate the unwanted signal frequencies from the For example, if in every state St, there exists an agent which
ones in interest. can take up an action At to get to a new state St+1. The
agent will gain the capacity to learn and interact in different
K. K-Means states by increasing the numerical long term reward for the
K-means is an unsupervised learning method that is used agent. Consider an example where if an athlete runs fast in a
for the clustering problem. The way it works is by using an 100m race, he wins the medal. The athlete is the agent, the
algorithm to locate a partition to minimize the error between environment is the field or ground, the reward is the medal,
a cluster’s empirical mean and points within. Using these K the state is the 100 m race. This is shown in Fig. 15
clusters, K-means tries to minimize the summation of the A typical reinforcement learning setup consists of possible
squared errors [61]. states denoted by S, A be the possible set of actions, a state
There are two commonly used methods for initialization: transition function given by Delta: S*A which tends to state
Forgy and Random Partition. With the Forgy method, K S, a reward function given by S*A*S which tends to R, and a
observations are chosen randomly from the data set. These policy pi which defines state which tends to action A. [106]
observations are then used as the initial means. For the The representation is shown below:
Random Partition method, each observation is first assigned
a random cluster. This is then updated as the initial mean is δ :S×A→S (14)
computed such that it is at the center of the cluster. r :S×A×S →R (15)
One of the advantages of K-means is its easy implemen-
tation of high computational speed given that K is relatively π:S→A (16)

TABLE VIII
L INEAR D ISCRIMINANT A NALYSIS

Author(s) Machine Learning Method Application Data Set

Kirar, J. S.,Agrawal, R. K. (2018 [70] LDA Feature selection and classifica- Public
tion for EEG using LDA
Yuan, S., Zhou, W., Chen, L. 2018 [121] LDA Siezure detection using Bayesian 21 patients
LDA.
Liu, Y. H., Huang, S., Huang, Y. D. 2017 [76] LDA Motor Imagery EEG classifica- Not Reported
tion
Neto, E., Biessmann, F., Aurlien, H., Nordby, H., Eichele, T. 2016 [84] LDA Regularized LDA of EEG fea- 114 Patients
tures
Treder, M. S., Porbadnigk, A. K., Avarvand, F. S., Müller, K. R., Blankertz, B. 2016 [112] LDA Optimal estimation of ERP Public
source time
Chen, W., Shen, C. P., Chiu, M. J., Zhao, Q., Cichocki, A., Lin, J. W., Lai, F. 2015 [19] LDA Epileptic EEG visualization 2 normal, 4 seizure patients
based on LDA
Mirsadeghi, M., Behnam, H., Shalbaf, R., Moghadam, H. J. 2016 [79] LDA Characterizing awake and anes- 25 Patients
thetized states using LDA
Ying, X., Lin, H., Hui, G. 2015 [119] LDA Non-linear bistable dynamics Public
model based on LDA
Onishi, A. and Natsume, K., 2014 [86] LDA Multi-class ERP based BCI anal- Not Reported
ysis
Onishi, A., Natsume, K. 2014, [86] LDA Epileptic seizure detection Public
Bayesian LDA

Fig. 14. General K-means classification. K-means works based on using an algorithm to locate a partition in order to minimize the error between a cluster’s
empirical mean and points within. Using these K clusters, K-means tries to minimize the summation of the squared errors

TABLE IX
K MEANS FOR EEG A NALYSIS

Author(s) Application Data Set Results

Manjusha, M., et al., 2016 [77] Robust Epilepsy Classification 20 patients Accuracy 93.02%
Prabhakar, et al., 2015 [89] Epilepsy Risk Level Classification 20 patients Accuracy 71.09%
Rai, et al., 2015 [93] Novel Feature Identification 5 patients Accuracy 99.00%
Teramae, et al., 2010 [110] estimation of feeling patients not listed discrimination ratio 84.2%
Harikumar, et al., 2012 [36] fuzzy outputs optimization 20 patients Accuracy 95.88%
Bizopoulos, et al., 2013 [17] epileptic seizure detection patients not listed Accuracy 98%
Asanza, et al., 2016 [12] EEG occipital signal classification patients not listed Accuracy unknown
ozerdem et al., 2017 [87] KNN,SVM Emotion recognition 32 healthy subjects 72.92%

for emotion recognition, EEG is one of the most reliable

because its signals are highly accurate and more objective than
other external appearance approaches.
The art of machine learning has led to the development and
application of different techniques, which has made it possible
for the computer to analyze and learn information from a given
set of data, and make the desired prediction accordingly. By
providing a sufficient amount of data, along with the help of
precision coding and algorithms, it has been possible to train
a computer in determining the exact output as required [49].
The proposed machine learning models are being used in
interpreting the type of epileptic disorder being treated, and
evaluate the treatment methods by analyzing EEG signals.
EEG signals play a major role in the accurate segregation of
different forms of epilepsy. The EEG signals are extracted
by the application of discrete wavelet transforms and auto-
regressive mathematical models and are supplied as input with
Fig. 15. Operation of Reinforcement Learning. Software agents must to take the ML algorithm. In this paper, we presented a broad and
suitable actions in an environment to maximize reward in a particular situation. comprehensive review of state-of-art EEG analysis methods
with medical applications. We tried to familiarize the readers
with machine learning methods which have been implemented
One of the advantages of using the RL model is that it
for different application.
maintains the balance between Exploration and Exploitation.
As the process of epileptic seizure detection is a bit com-
Other supervised algorithms cannot perform this balance. For
plicated biomedical situation, it has generated a substantial
EEG analysis applications, the RL model has shown constant
amount of concerns towards the utilization of machine learning
progress towards the control mechanism of the brain-computer
processes as a solution [83]. Most of the recent literature
interface system, maintaining an equal balance between state
surveys regarding EEG signal analysis have proposed mul-
transitions and reward mechanisms for optimum functioning.
tiple learning models and different artificial neural network
algorithms like radial bias function, recurrent neural networks,
M. Combination of Methods and vector quantizations to interpret epileptic seizure patterns
A combination of methods involves the use of two or more in a given set of EEG signals. The problem is also being
of the machine learning algorithms to take advantage of the targeted and solved using other models like Support Vector
unique characteristics that each method possesses. This allows Machines (SVM), adaptive neuro-fuzzy interference system
the multimodal algorithm to extract additional desired features. (ANFIS), adaptive learning, and time-frequency analysis.
The significance of multimodal integration is that it allows Reviewing the published papers in EEG analysis for
high-resolution classification using primarily already existing epilepsy the following points are considerable. Dimensionality
methods [23], [50]. Additionally, this resolution will generally reduction and selection have been identified as an interesting
be higher than that of the individual methods separately. How- topic for EEG analysis using machine learning methods [13].
ever, multimodal extraction is not without limitations. Due to Also, wavelet transform and Auto-regressive methods have
the increased complexity of the algorithm, it may be difficult played a pivotal role in machine learning for EEG such as the
to determine the true accuracy as it is not directly comparable following studies [7]. Subasi used wavelet feature extraction
to existing methods. An example of this application in EEG is for epileptic seizure detection with an adaptive neuro-fuzzy
the diagnosis of multiple sclerosis patients. In the paper, the T- inference system in‘ [109]. Linear programming boosting has
test [45], [46] and Bhattacharyya were used for feature extrac- been used for EEG signals analysis in [37]. The effect of
tion as part of the preprocessing. Following this a combination de-noising such as multiscale PCA in EEG analysis is shown
of KNN and SVM as the primary classification algorithm. in [67]. Data preparation methods such as PCA, ICA and,
This resulted in a total accuracy of 93% [111]. While other LDA can be used to increase the classification accuracy [34].
sections above have dedicated tables with reviewed literature, Ensemble methods and combining classifiers have shown good
we wanted to bring attention to multimodal analysis as some performance in EEG analysis such as the following studies
literature above already demonstrated the application of the [66].
combination of methods [41]. The incorporation of deep learning models in neuroimaging
and electrodiagnostic analytics has allowed for large amounts
III. C ONCLUSION of data to be correlated from multiple modalities [44]. These
EEG is a noninvasive electrophysiological device that models have been shown to perform better and faster than
records the electrical activity of the brain by placing electrodes current state-of-the-art analysis techniques through both super-
around the scalp. Based on this activity and oscillating electric vised and unsupervised learning tasks. Recent advancements
potentials, EEG can be used to diagnose neurological disor- and advantages in using deep learning in EEG analysis can
ders, such as epilepsy, or for emotion recognition. Of methods provide more accurate and faster analysis for a large amount

of data. Hosseini et al. [39], [42], [51] proposed a cloud- [2] H. Abbasi, L. Bennet, A. J. Gunn, and C. P. Unsworth. Identify-
based method for EEG analysis. In [52] convolutional neural ing stereotypic evolving micro-scale seizures (sems) in the hypoxic-
ischemic eeg of the pre-term fetal sheep with a wavelet type-ii fuzzy
networks (CNN) have been developed for EEG analysis. classifier. In Engineering in Medicine and Biology Society (EMBC),
In [47] optimization modules consisted of PCA, ICA and DSA 2016 IEEE 38th Annual International Conference of the, pages 973–
analysis are developed for CNN and stacked auto-encoder deep 976. IEEE, 2016.
[3] H. Aghajani, M. Garbey, and A. Omurtag. Measuring mental workload
learning structures in EEG analysis. with eeg+ fnirs. Frontiers in human neuroscience, 11:359, 2017.
We reviewed different applications of machine learning [4] A. Ahani, H. Wahbeh, H. Nezamfar, M. Miller, D. Erdogmus, and
in EEG such as sleep stage analysis. Alickovic et al. [8] B. Oken. Quantitative change of eeg and respiration signals during
mindfulness meditation. Journal of neuroengineering and rehabilita-
and Hassan et al. [38] proposed automated identification of tion, 11(1):87, 2014.
sleep stages from single-channel EEG signals. Kevric et al. [5] S. B. Akben, A. Subasi, and D. Tuncel. Analysis of repetitive flash
presented signal decomposition methods in the classification stimulation frequencies and record periods to detect migraine using
artificial neural network. Journal of medical systems, 36(2):925–931,
of EEG signals for the motor-imagery BCI system. 2012.
The coefficients of the wavelet transform and the numerical [6] O. Al Zoubi, C. Ki Wong, R. T. Kuplicki, H.-w. Yeh, A. Mayeli,
autoregressive model are used in recognizing the changes and H. Refai, M. Paulus, and J. Bodurka. Predicting age from brain eeg
signals–a machine learning approach. Frontiers in aging neuroscience,
behaviors in EEG signals. These coefficients are taken as in- 10:184, 2018.
puts and combined with different machine learning algorithms [7] E. Alickovic, J. Kevric, and A. Subasi. Performance evaluation of
like multiple layered neural networks, K-means, Support vec- empirical mode decomposition, discrete wavelet transform, and wavelet
packed decomposition for automated epileptic seizure detection and
tor machines, K-nearest neighbors, and Naive Bayesian; to prediction. Biomedical signal processing and control, 39:94–102, 2018.
break the EEG signal into machine recognizable components, [8] E. Alickovic and A. Subasi. Ensemble svm method for automatic
for extracting and determining the power points which are sleep stage classification. IEEE Transactions on Instrumentation and
Measurement, 67(6):1258–1265, 2018.
responsible for triggering seizures. [9] H. U. Amin, W. Mumtaz, A. R. Subhani, M. N. M. Saad, and A. S.
As there are multiple techniques involving machine learning Malik. Classification of eeg signals based on pattern recognition
for analyzing a given set of EEG signals, it is required to approach. Frontiers in computational neuroscience, 11:103, 2017.
[10] M. N. Anastasiadou, M. Christodoulakis, E. S. Papathanasiou, S. S.
evaluate the best-suited technique for a given application. Each Papacostas, and G. D. Mitsis. Unsupervised detection and removal of
model has a specific use case about the type of application muscle artifacts from scalp eeg recordings using canonical correlation
and subject data set. As to our topic of study here, we were analysis, wavelets and random forests. Clinical Neurophysiology,
128(9):1755–1769, 2017.
concerned about the analysis of waveforms to determine an [11] A. Antoniades, L. Spyrou, D. Martin-Lopez, A. Valentin, G. Alarcon,
output. Here we will see how each different ML models can S. Sanei, and C. C. Took. Deep neural architectures for mapping
be used for the intended use case: scalp to intracranial eeg. International journal of neural systems, page
1850009, 2018.
K-NN classifiers can be used for both regression and [12] V. Asanza, K. Ochoa, C. Sacarelo, C. Salazar, F. Loayza, C. Vaca,
classification of data, which for our purpose can be used for and E. Peláez. Clustering of eeg occipital signals using k-means. In
identifying and classifying different acquired EEG signals and Ecuador Technical Chapters Meeting (ETCM), IEEE, pages 1–5. IEEE,
2016.
finding the nearest possible output point to the desired classifi- [13] N. Beganovic, J. Kevric, and D. Jokic. Identification of diagnostic-
cation line for possible detection of abnormality. ANN, on the related features applicable to eeg signal analysis. In Proceedings of
other hand, has the capability of segregating the physical shape the Annual Conference of the PHM Society, volume 10, 2018.
[14] M. Bentlemsan, E.-T. Zemouri, D. Bouchaffra, B. Yahya-Zoubir, and
of the EEG waveform and dividing it into segments. These K. Ferroudji. Random forest and filter bank common spatial patterns
segments are each given a specific weight value accordingly for eeg-based motor imagery classification. In Intelligent Systems,
by analyzing the waveform, and the output is determined by Modelling and Simulation (ISMS), 2014 5th International Conference
on, pages 235–238. IEEE, 2014.
subjecting the final equation to a bias. The final chosen bias
[15] N. Bigdely-Shamlo, A. Vankov, R. R. Ramirez, and S. Makeig. Brain
brings down the output to a desired expected range. As more activity-based image classification from rapid serial visual presentation.
data is being involved, the number of interactions in the hidden IEEE Transactions on Neural Systems and Rehabilitation Engineering,
layer will increase. So depending on the type of problem and 16(5):432–441, 2008.
[16] S. Biswal, Z. Nip, V. M. Junior, M. T. Bianchi, E. S. Rosenthal, and
the amount of data being considered, a suitable selection has M. B. Westover. Automated information extraction from free-text eeg
to be made while selecting an appropriate process. reports. In Engineering in Medicine and Biology Society (EMBC), 2015
37th Annual International Conference of the IEEE, pages 6804–6807.
IEEE, 2015.
[17] P. A. Bizopoulos, D. G. Tsalikakis, A. T. Tzallas, D. D. Koutsouris,
IV. ACKNOWLEDGEMENT and D. I. Fotiadis. Eeg epileptic seizure detection using k-means
clustering and marginal spectrum based on ensemble empirical mode
We would like to thank students at Santa Clara University, decomposition. In Bioinformatics and Bioengineering (BIBE), 2013
IEEE 13th International Conference on, pages 1–4. IEEE, 2013.
Thi-Vu Huynh, Pradnya Patel, Elissa Yang, and Haygreev [18] S. Bose, V. Rama, and C. R. Rao. Eeg signal analysis for seizure
Venkatesh for their contributions to this work. detection using discrete wavelet transform and random forest. In
Computer and Applications (ICCA), 2017 International Conference on,
pages 369–378. IEEE, 2017.
[19] W. Chen, C.-P. Shen, M.-J. Chiu, Q. Zhao, A. Cichocki, J.-W. Lin,
R EFERENCES and F. Lai. Epileptic eeg visualization and sonification based on linear
discriminate analysis. In Engineering in Medicine and Biology Society
[1] A. Aarabi, R. Fazel-Rezai, and Y. Aghakhani. Seizure detection in (EMBC), 2015 37th Annual International Conference of the IEEE,
intracranial eeg using a fuzzy inference system. In Engineering in pages 4466–4469. IEEE, 2015.
Medicine and Biology Society, 2009. EMBC 2009. Annual International [20] A. M. Chiarelli, P. Croce, A. Merla, and F. Zappasodi. Deep learning
Conference of the IEEE, pages 1860–1863. IEEE, 2009. for hybrid eeg-fnirs brain–computer interface: application to motor

imagery classification. Journal of neural engineering, 15(3):036028, [40] M.-P. Hosseini, A. Hajisami, and D. Pompili. Real-time epileptic
2018. seizure detection from eeg signals via random subspace ensemble learn-
[21] E. Combrisson and K. Jerbi. Exceeding chance level by chance: The ing. In 2016 IEEE international conference on autonomic computing
caveat of theoretical chance levels in brain signal classification and (ICAC), pages 209–218. IEEE, 2016.
statistical assessment of decoding accuracy. Journal of neuroscience [41] M.-P. Hosseini, A. Lau, K. Elisevich, and H. Soltanian-Zadeh. Mul-
methods, 250:126–136, 2015. timodal analysis in biomedicine. In Big Data in Multimodal Medical
[22] M. Cosenza-Andraus, C. Nunes-Cosenza, R. Gomes- Imaging, pages 193–203. Chapman and Hall/CRC, 2019.
Nunes, C. Fantezia-Andraus, and S. Alves-Leon. Video- [42] M.-P. Hosseini, S. Lu, K. Kamaraj, A. Slowikowski, and H. C.
electroencephalography prolonged monitoring in patients with Venkatesh. Deep learning architectures. In Deep Learning: Concepts
ambulatory diagnosis of medically refractory temporal lobe epilepsy: and Architectures, pages 1–24. Springer, 2020.
application of fuzzy logic’s model. Revista de neurologia, 43(1):7–14, [43] M.-P. Hosseini, M. R. Nazem-Zadeh, F. Mahmoudi, H. Ying, and
2006. H. Soltanian-Zadeh. Support vector machine with nonlinear-kernel
[23] S. Dähne, F. Bießmann, F. C. Meinecke, J. Mehnert, S. Fazli, and optimization for lateralization of epileptogenic hippocampus in mr
K.-R. Mtüller. Multimodal integration of electrophysiological and images. In 2014 36th Annual International Conference of the IEEE
hemodynamic signals. In Brain-Computer Interface (BCI), 2014 Engineering in Medicine and Biology Society, pages 1047–1050. IEEE,
International Winter Workshop on, pages 1–4. IEEE, 2014. 2014.
[24] A. Delorme and S. Makeig. Eeglab: an open source toolbox for analysis [44] M.-P. Hosseini, M.-R. Nazem-Zadeh, D. Pompili, K. Jafari-Khouzani,
of single-trial eeg dynamics including independent component analysis. K. Elisevich, and H. Soltanian-Zadeh. Automatic and manual seg-
Journal of neuroscience methods, 134(1):9–21, 2004. mentation of hippocampus in epileptic patients mri. arXiv preprint
[25] J. A. Dian, S. Colic, Y. Chinvarun, P. L. Carlen, and B. L. Bardakjian. arXiv:1610.07557, 2016.
Identification of brain regions of interest for epilepsy surgery planning [45] M.-P. Hosseini, M.-R. Nazem-Zadeh, D. Pompili, K. Jafari-Khouzani,
using support vector machines. In Engineering in Medicine and Biology K. Elisevich, and H. Soltanian-Zadeh. Comparative performance
Society (EMBC), 2015 37th Annual International Conference of the evaluation of automated segmentation methods of hippocampus from
IEEE, pages 6590–6593. IEEE, 2015. magnetic resonance images of temporal lobe epilepsy patients. Medical
[26] Q. Dong, B. Hu, J. Zhang, X. Li, and M. Ratcliffe. A study on visual physics, 43(1):538–553, 2016.
attention modeling—a linear regression method based on eeg. In Neural [46] M.-P. Hosseini, M. R. Nazem-Zadeh, D. Pompili, and H. Soltanian-
Networks (IJCNN), The 2013 International Joint Conference on, pages Zadeh. Statistical validation of automatic methods for hippocampus
1–6. IEEE, 2013. segmentation in mr images of epileptic patients. In 2014 36th Annual
[27] C. Dora and P. K. Biswal. Robust ecg artifact removal from eeg International Conference of the IEEE Engineering in Medicine and
using continuous wavelet transformation and linear regression. In Biology Society, pages 4707–4710. IEEE, 2014.
Signal Processing and Communications (SPCOM), 2016 International [47] M.-P. Hosseini, D. Pompili, K. Elisevich, and H. Soltanian-Zadeh.
Conference on, pages 1–5. IEEE, 2016. Optimized deep learning for eeg big data and seizure prediction bci
[28] F. D. V. Fallani, G. Vecchiato, J. Toppi, L. Astolfi, and F. Babiloni. via internet of things. IEEE Transactions on Big Data, 3(4):392–404,
Subject identification through standard eeg signals during resting states. 2017.
In 2011 Annual International Conference of the IEEE Engineering in [48] M.-P. Hosseini, D. Pompili, K. Elisevich, and H. Soltanian-Zadeh.
Medicine and Biology Society, pages 2331–2333. IEEE, 2011. Random ensemble learning for eeg classification. Artificial intelligence
in medicine, 84:146–158, 2018.
[29] J. Fan, J. W. Wade, D. Bian, A. P. Key, Z. E. Warren, L. C. Mion,
[49] M. P. Hosseini, H. Soltanian-Zadeh, and S. Akhlaghpoor. Computer-
and N. Sarkar. A step towards eeg-based brain computer interface
aided diagnosis system for the evaluation of chronic obstructive pul-
for autism intervention. In Conference proceedings:... Annual Interna-
monary disease on ct images. Tehran University Medical Journal,
tional Conference of the IEEE Engineering in Medicine and Biology
68(12), 2011.
Society. IEEE Engineering in Medicine and Biology Society. Annual
[50] M. P. Hosseini, H. Soltanian-Zadeh, and S. Akhlaghpoor. Three cuts
Conference, volume 2015, page 3767. NIH Public Access, 2015.
method for identification of copd. Acta Medica Iranica, pages 771–778,
[30] V. Gao, F. Turek, and M. Vitaterna. Multiple classifier systems for
2013.
automatic sleep scoring in mice. Journal of neuroscience methods,
[51] M.-P. Hosseini, H. Soltanian-Zadeh, K. Elisevich, and D. Pompili.
264:33–39, 2016.
Cloud-based deep learning of big eeg data for epileptic seizure pre-
[31] S. Ghosh-Dastidar and H. Adeli. A new supervised learning algorithm diction. In 2016 IEEE global conference on signal and information
for multiple spiking neural networks with application in epilepsy and processing (GlobalSIP), pages 1151–1155. IEEE, 2016.
seizure detection. Neural networks, 22(10):1419–1431, 2009. [52] M.-P. Hosseini, T. X. Tran, D. Pompili, K. Elisevich, and H. Soltanian-
[32] M. Günay and T. Ensari. Eeg signal analysis of patients with epilepsy Zadeh. Deep learning with edge computing for localization of epilep-
disorder using machine learning techniques. In 2018 Electric Elec- togenicity using multimodal rs-fmri and eeg big data. In 2017 IEEE
tronics, Computer Science, Biomedical Engineerings’ Meeting (EBBT), International Conference on Autonomic Computing (ICAC), pages 83–
pages 1–4. IEEE, 2018. 92. IEEE, 2017.
[33] L. Guo, D. Rivero, and A. Pazos. Epileptic seizure detection using [53] A. E. Hramov, V. A. Maksimenko, S. V. Pchelintseva, A. E. Runnova,
multiwavelet transform based approximate entropy and artificial neural V. V. Grubov, V. Y. Musatov, M. O. Zhuravlev, A. A. Koronovskii,
networks. Journal of neuroscience methods, 193(1):156–163, 2010. and A. N. Pisarchik. Classifying the perceptual interpretations of a
[34] M. I. Gursoy and A. Subast. A comparison of pca, ica and lda in eeg bistable image using eeg and artificial neural networks. Frontiers in
signal classification using svm. In 2008 Ieee 16th Signal Processing, neuroscience, 11:674, 2017.
Communication and Applications Conference, pages 1–4. IEEE, 2008. [54] W.-Y. Hsu. Assembling a multi-feature eeg classifier for left–right
[35] C. R. Hamilton, S. Shahryari, and K. M. Rasheed. Eye state prediction motor imagery data using wavelet-based fuzzy approximate entropy
from eeg data using boosted rotational forests. In Machine Learning for improved accuracy. International journal of neural systems,
and Applications (ICMLA), 2015 IEEE 14th International Conference 25(08):1550037, 2015.
on, pages 429–432. IEEE, 2015. [55] J. Hu and J. Min. Automated detection of driver fatigue based on
[36] R. Harikumar, T. Vijayakumar, and M. Sreejith. Performance analysis eeg signals using gradient boosting decision tree model. Cognitive
of svd and k-means clustering for optimization of fuzzy outputs in Neurodynamics, pages 1–10, 2018.
classification of epilepsy risk level from eeg signals. In Electrical En- [56] J. Hu and Z. Mu. Eeg authentication system based on auto-regression
gineering/Electronics, Computer, Telecommunications and Information coefficients. In Intelligent Systems and Control (ISCO), 2016 10th
Technology (ECTI-CON), 2012 9th International Conference on, pages International Conference on, pages 1–5. IEEE, 2016.
1–4. IEEE, 2012. [57] Y. Huang, J. Yang, P. Liao, and J. Pan. Fusion of facial expressions and
[37] A. R. Hassan and A. Subasi. Automatic identification of epileptic eeg for multimodal emotion recognition. Computational intelligence
seizures from eeg signals using linear programming boosting. Com- and neuroscience, 2017, 2017.
puter methods and programs in biomedicine, 136:65–77, 2016. [58] A. Ishfaque, A. J. Awan, N. Rashid, and J. Iqbal. Evaluation of ann,
[38] A. R. Hassan and A. Subasi. A decision support system for auto- lda and decision trees for eeg based brain computer interface. In
mated identification of sleep stages from single-channel eeg signals. Proceedings of the IEEE 9th International Conference on Emerging
Knowledge-Based Systems, 128:115–124, 2017. Technologies (ICET’13), pages 1–6, 2013.
[39] M. P. Hosseini. Brain-computer interface for analyzing epileptic big [59] I. Iturrate, L. Montesano, and J. Minguez. Robot reinforcement
data. PhD thesis, Rutgers University-School of Graduate Studies, 2018. learning using eeg-based reward signals. In Proceedings of the IEEE

International Conference on Robotics and Automation (ICRA), number [80] W. Mumtaz, S. S. A. Ali, M. A. M. Yasin, and A. S. Malik. A machine
EPFL-CONF-205134, pages 4822–4829, 2010. learning framework involving eeg-based functional connectivity to
[60] A. Jain, B. Abbas, O. Farooq, and S. K. Garg. Fatigue detection and diagnose major depressive disorder (mdd). Medical & biological
estimation using auto-regression analysis in eeg. In Advances in Com- engineering & computing, 56(2):233–246, 2018.
puting, Communications and Informatics (ICACCI), 2016 International [81] W. Mumtaz, N. Kamel, S. S. A. Ali, A. S. Malik, et al. An eeg-based
Conference on, pages 1092–1095. IEEE, 2016. functional connectivity measure for automatic detection of alcohol use
[61] A. K. Jain. Data clustering: 50 years beyond k-means. Pattern disorder. Artificial intelligence in medicine, 84:79–89, 2018.
recognition letters, 31(8):651–666, 2010. [82] M. Murakami, S. Nakatani, N. Araki, Y. Konishi, and K. Mabuchi.
[62] A. K. Jaiswal and H. Banka. Epileptic seizure detection in eeg signal Motion discrimination from eeg using logistic regression and schmitt-
with gmodpca and support vector machine. Bio-medical materials and trigger-type threshold. In Systems, Man, and Cybernetics (SMC), 2015
engineering, 28(2):141–157, 2017. IEEE International Conference on, pages 2338–2342. IEEE, 2015.
[63] L. Jakaite, V. Schetinin, C. Maple, and J. Schult. Bayesian decision [83] M.-R. Nazem-Zadeh, J. M. Schwalb, H. Bagher-Ebadian, F. Mahmoudi,
trees for eeg assessment of newborn brain maturity. In Computational M.-P. Hosseini, K. Jafari-Khouzani, K. V. Elisevich, and H. Soltanian-
Intelligence (UKCI), 2010 UK Workshop on, pages 1–6. IEEE, 2010. Zadeh. Lateralization of temporal lobe epilepsy by imaging-based
[64] A. Jalilifard, E. B. Pizzolato, and M. K. Islam. Emotion classification response-driven multinomial multivariate models. In 2014 36th Annual
using single-channel scalp-eeg recording. In Engineering in Medicine International Conference of the IEEE Engineering in Medicine and
and Biology Society (EMBC), 2016 IEEE 38th Annual International Biology Society, pages 5595–5598. IEEE, 2014.
Conference of the, pages 845–849. IEEE, 2016. [84] E. Neto, F. Biessmann, H. Aurlien, H. Nordby, and T. Eichele.
[65] S. Jirayucharoensak, S. Pan-Ngum, and P. Israsena. Eeg-based emotion Regularized linear discriminant analysis of eeg features in dementia
recognition using deep learning network with principal component patients. Frontiers in aging neuroscience, 8:273, 2016.
based covariate shift adaptation. The Scientific World Journal, 2014, [85] S. N. Oğulata, C. Şahin, and R. Erol. Neural network-based computer-
2014. aided diagnosis in classification of primary generalized epilepsy by eeg
[66] S. Jukić and J. Kevrić. Majority vote of ensemble machine learning signals. Journal of medical systems, 33(2):107–112, 2009.
methods for real-time epilepsy prediction applied on eeg pediatric data. [86] A. Onishi and K. Natsume. Multi-class erp-based bci data analysis
TEM Journal, 7(2):313, 2018. using a discriminant space self-organizing map. In Engineering in
[67] J. Kevric and A. Subasi. The effect of multiscale pca de-noising in Medicine and Biology Society (EMBC), 2014 36th Annual International
epileptic seizure detection. Journal of medical systems, 38(10):131, Conference of the IEEE, pages 26–29. IEEE, 2014.
2014. [87] M. S. Özerdem and H. Polat. Emotion recognition based on eeg
[68] J.-H. Kim, F. Bießmann, and S.-W. Lee. Reconstruction of hand features in movie clips with channel selection. Brain informatics,
movements from eeg signals based on non-linear regression. In Brain- 4(4):241, 2017.
Computer Interface (BCI), 2014 International Winter Workshop on, [88] A. Page, S. P. T. Oates, and T. Mohsenin. An ultra low power feature
pages 1–3. IEEE, 2014. extraction and classification system for wearable seizure detection.
[69] C. Kingsford and S. L. Salzberg. What are decision trees? Nature In Engineering in Medicine and Biology Society (EMBC), 2015 37th
biotechnology, 26(9):1011, 2008. Annual International Conference of the IEEE, pages 7111–7114. IEEE,
[70] J. S. Kirar and R. Agrawal. Relevant feature selection from a 2015.
combination of spectral-temporal and spatial features for classification [89] S. K. Prabhakar and H. Rajaguru. Pca and k-means clustering for
of motor imagery eeg. Journal of medical systems, 42(5):78, 2018. classification of epilepsy risk levels from eeg signals? a comparitive
[71] J. Laton, J. Van Schependom, J. Gielen, J. Decoster, T. Moons, study between them. In 2015 International Conference on Intelligent
J. De Keyser, M. De Hert, and G. Nagels. Single-subject classification Informatics and Biomedical Sciences (ICIIBMS), pages 83–86. IEEE,
of schizophrenia patients based on a combination of oddball and 2015.
mismatch evoked potential paradigms. Journal of the neurological [90] S. Puntanen. Regression analysis by example, by samprit chatterjee,
sciences, 347(1-2):262–267, 2014. ali s. hadi. International Statistical Review, 81(2):308–308, 2013.
[72] J. Le Douget, A. Fouad, M. M. Filali, J. Pyrzowski, and [91] A. F. Rabbi, L. Azinfar, and R. Fazel-Rezai. Seizure prediction using
M. Le Van Quyen. Surface and intracranial eeg spike detection based adaptive neuro-fuzzy inference system. In Engineering in Medicine and
on discrete wavelet decomposition and random forest classification. In Biology Society (EMBC), 2013 35th Annual International Conference
2017 39th Annual International Conference of the IEEE Engineering of the IEEE, pages 2100–2103. IEEE, 2013.
in Medicine and Biology Society (EMBC), pages 475–478. IEEE, 2017. [92] A. F. Rabbi and R. Fazel-Rezai. A fuzzy logic system for seizure
[73] Y.-H. Lee, Y.-J. Hsieh, Y.-J. Shiah, Y.-H. Lin, C.-Y. Chen, Y.-C. Tyan, onset detection in intracranial eeg. Computational intelligence and
J. GengQiu, C.-Y. Hsu, and S. C.-J. Chen. A cross-sectional evaluation neuroscience, 2012:1, 2012.
of meditation experience on electroencephalography data by artificial [93] K. Rai, V. Bajaj, and A. Kumar. Novel feature for identification of
neural network and support vector machine classifiers. Medicine, focal eeg signals with k-means and fuzzy c-means algorithms. In
96(16), 2017. Digital Signal Processing (DSP), 2015 IEEE International Conference
[74] P. Li, C. Karmakar, J. Yearwood, S. Venkatesh, M. Palaniswami, and on, pages 412–416. IEEE, 2015.
C. Liu. Detection of epileptic seizure based on entropy analysis of [94] H. Rajaguru and S. K. Prabhakar. Logistic regression gaussian mixture
short-term eeg. PloS one, 13(3):e0193691, 2018. model and softmax discriminant classifier for epilepsy classification
[75] X. Li, C. Guan, K. K. Aug, C. Wang, Z. Y. Chin, H. Zhang, C. G. from eeg signals. In Computing Methodologies and Communication
Lim, and T. S. Lee. An ocular artefacts correction method for (ICCMC), 2017 International Conference on, pages 985–988. IEEE,
discriminative eeg analysis based on logistic regression. In Signal 2017.
Processing Conference (EUSIPCO), 2015 23rd European, pages 2731– [95] H. Rajaguru and S. K. Prabhakar. Non linear ica and logistic regression
2735. IEEE, 2015. for classification of epilepsy from eeg signals. In Electronics, Com-
[76] Y.-H. Liu, S. Huang, and Y.-D. Huang. Motor imagery eeg classi- munication and Aerospace Technology (ICECA), 2017 International
fication for patients with amyotrophic lateral sclerosis using fractal conference of, volume 1, pages 577–580. IEEE, 2017.
dimension and fisher’s criterion-based channel selection. Sensors, [96] H. Rajaguru and S. K. Prabhakar. Sparse pca and soft decision tree
17(7):1557, 2017. classifiers for epilepsy classification from eeg signals. In Electronics,
[77] M. Manjusha and R. Harikumar. Performance analysis of knn classifier Communication and Aerospace Technology (ICECA), 2017 Interna-
and k-means clustering for robust classification of epilepsy from tional conference of, volume 1, pages 581–584. IEEE, 2017.
eeg signals. In Wireless Communications, Signal Processing and [97] I. Rish et al. An empirical study of the naive bayes classifier. In
Networking (WiSPNET), International Conference on, pages 2412– IJCAI 2001 workshop on empirical methods in artificial intelligence,
2416. IEEE, 2016. volume 3, pages 41–46. IBM New York, 2001.
[78] T. Meyer, J. Peters, T. O. Zander, B. Schölkopf, and M. Grosse- [98] S. Roy, I. Kiral-Kornek, and S. Harrer. Deep learning enabled auto-
Wentrup. Predicting motor learning performance from electroen- matic abnormal eeg identification. In 2018 40th Annual International
cephalographic data. Journal of neuroengineering and rehabilitation, Conference of the IEEE Engineering in Medicine and Biology Society
11(1):24, 2014. (EMBC), pages 2756–2759. IEEE, 2018.
[79] M. Mirsadeghi, H. Behnam, R. Shalbaf, and H. J. Moghadam. Charac- [99] K.-M. Rytkönen, J. Zitting, and T. Porkka-Heiskanen. Automated sleep
terizing awake and anesthetized states using a dimensionality reduction scoring in rats and mice using the naive bayes classifier. Journal of
method. Journal of medical systems, 40(1):13, 2016. neuroscience methods, 202(1):60–64, 2011.

[100] C. Y. Sai, N. Mokhtar, H. Arof, P. Cumming, and M. Iwahashi. [121] S. Yuan, W. Zhou, and L. Chen. Epileptic seizure prediction using dif-
Automated classification and removal of eeg artifacts with svm and fusion distance and bayesian linear discriminate analysis on intracranial
wavelet-ica. IEEE journal of biomedical and health informatics, eeg. International journal of neural systems, 28(01):1750043, 2018.
22(3):664–670, 2018. [122] T. Zhang and W. Chen. Lmd based features for the automatic seizure
[101] J. Saini and M. Dutta. Epilepsy classification using optimized artificial detection of eeg signals using svm. IEEE Transactions on Neural
neural network. Neurological research, pages 1–13, 2018. Systems and Rehabilitation Engineering, 25(8):1100–1108, 2017.
[102] B. Sharif and A. H. Jafari. Prediction of epileptic seizures from eeg [123] N. Zhuang, Y. Zeng, K. Yang, C. Zhang, L. Tong, and B. Yan.
using analysis of ictal rules on poincaré plane. Computer methods and Investigating patterns for self-induced emotion recognition from eeg
programs in biomedicine, 145:11–22, 2017. signals. Sensors, 18(3):841, 2018.
[103] A. Sharma, J. Rai, and R. Tewari. Epileptic seizure anticipation and
localisation of epileptogenic region using eeg signals. Journal of
Medical Engineering & Technology, pages 1–14, 2018.
[104] A. Sharmila and P. Geethanjali. Effect of filtering with time domain
features for the detection of epileptic seizure from eeg signals. Journal
of Medical Engineering & Technology, pages 1–11, 2018.
[105] A. Sharmila and P. Mahalakshmi. Wavelet-based feature extraction
for classification of epileptic seizure eeg signal. Journal of medical
engineering & technology, 41(8):670–680, 2017. Mohammad-Parsa Hosseini is a senior member
[106] E. Shephard, G. M. Jackson, and M. J. Groom. Electrophysiological at IEEE. He received his PhD in Electrical and
correlates of reinforcement learning in young people with tourette Computer Engineering (ECE) with research in Com-
syndrome with and without co-occurring adhd symptoms. International puter Science from Rutgers University, NJ, USA in
Journal of Developmental Neuroscience, 51:17–27, 2016. 2018. He had previously received his BSc degree
in Electrical and Electronic Engineering in 2006, a
[107] V. Srinivasan, C. Eswaran, and N. Sriraam. Approximate entropy-
graduate study in MSc of Biomedical Engineering
based epileptic eeg detection using artificial neural networks. IEEE
in 2008, and a second MSc program in Electrical
Transactions on information Technology in Biomedicine, 11(3):288–
and Communication Engineering in 2010. He is
295, 2007.
collaborating with Medical Image Analysis Lab at
[108] A. F. Struck, B. Ustun, A. R. Ruiz, J. W. Lee, S. M. LaRoche,
Henry Ford Health System and with the Clinical
L. J. Hirsch, E. J. Gilmore, J. Vlachy, H. A. Haider, C. Rudin, et al.
Neurosciences Department, Spectrum Health, MI, USA. He is working as
Association of an electroencephalography-based risk score with seizure
a data scientist and machine learning researcher at Silicon Valley, CA, USA
probability in hospitalized patients. JAMA neurology, 74(12):1419–
since 2017. He has been teaching as an adjunct lecture and faculty member at
1424, 2017.
several universities since 2009 and currently is with Santa Clara University.
[109] A. Subasi. Application of adaptive neuro-fuzzy inference system for His current research focus in machine learning, deep learning, signal and
epileptic seizure detection using wavelet feature extraction. Computers image processing. He has served on the scientific committees and review
in biology and medicine, 37(2):227–244, 2007. boards of several national and international conferences and journals.
[110] T. Teramae, D. Kushida, F. Takemori, and A. Kitamura. Estimation of
feeling based on eeg by using nn and k-means algorithm for massage
system. In SICE Annual Conference 2010, Proceedings of, pages 1542–
1547. IEEE, 2010.
[111] A. Torabi, M. R. Daliri, and S. H. Sabzposhan. Diagnosis of multiple
sclerosis from eeg signals using nonlinear methods. Australasian
physical & engineering sciences in medicine, 40(4):785–797, 2017.
[112] M. S. Treder, A. K. Porbadnigk, F. S. Avarvand, K.-R. Müller, and
B. Blankertz. The lda beamformer: optimal estimation of erp source Amin Hosseini is an IEEE member in signal pro-
time series using linear discriminant analysis. Neuroimage, 129:279– cessing and machine learning societies. He is with
291, 2016. the Electrical and Computer Engineering with minor
[113] V. Tuyisenge, L. Trebaul, M. Bhattacharjee, B. Chanteloup-Forêt, in computer science at Azad University, Tehran
C. Saubat-Guigui, I. Mîndruţă, S. Rheims, L. Maillard, P. Kahane, Central Branch. His research interests include digi-
D. Taussig, et al. Automatic bad channel detection in intracranial tal signal and image processing, machine learning,
electroencephalographic recordings using ensemble machine learning. artificial intelligence and biomedical engineering.
Clinical Neurophysiology, 129(3):548–554, 2018.
[114] A. T. Tzallas, M. G. Tsipouras, and D. I. Fotiadis. Automatic
seizure detection based on time-frequency analysis and artificial neural
networks. Computational Intelligence and Neuroscience, 2007, 2007.
[115] E. D. Ubeyli. Fuzzy similarity index for discrimination of eeg signals.
In Engineering in Medicine and Biology Society, 2006. EMBS’06. 28th
Annual International Conference of the IEEE, pages 5346–5349. IEEE,
2006.
[116] V. Vijayakumar, M. Case, S. Shirinpour, and B. He. Quantifying and
characterizing tonic thermal pain across subjects from eeg data using
random forest models. IEEE Transactions on Biomedical Engineering,
Kiarash Ahi received his M.Sc. degree in Electrical
64(12):2988–2996, 2017.
and Information Engineering from the Leibniz Uni-
[117] Y. Wang, W. Chen, K. Huang, and Q. Gu. Classification of neonatal
versity of Hannover, Germany in 2012, and his Ph.D.
amplitude-integrated eeg using random forest model with combined
degree in Electrical and Computer Engineering from
feature. In Bioinformatics and Biomedicine (BIBM), 2013 IEEE
the University of Connecticut, USA in 2017. He is
International Conference On, pages 285–290. IEEE, 2013.
currently a Senior Researcher and Lead Product De-
[118] S. Weichwald, T. Meyer, B. Scholkopf, T. Ball, and M. Grosse-
velopment Engineer in the advanced Semiconductor
Wentrup. Decoding index finger position from eeg using random
Industry. He architects automated systems, empow-
forests. In Cognitive Information Processing (CIP), 2014 4th Inter-
ered by Machine Learning and Image Processing,
national Workshop on, pages 1–6. IEEE, 2014.
and leads multinational RD teams. His research
[119] X. Ying, H. Lin, and G. Hui. Study on non-linear bistable dynamics and scientific interests include digital image and
model based eeg signal discrimination analysis method. Bioengineered, signal processing, semiconductor devices, Machine Learning, and intelligent
6(5):297–298, 2015. software development.
[120] W. Yu, T. Liu, R. Valdez, M. Gwinn, and M. J. Khoury. Application of
support vector machine modeling for prediction of common diseases:
the case of diabetes and pre-diabetes. BMC Medical Informatics and
Decision Making, 10(1):16, Mar 2010.

Presentation On ADS Momentum
No ratings yet
Presentation On ADS Momentum
345 pages
VITROS ValidationGuide 2020 PR-04963
No ratings yet
VITROS ValidationGuide 2020 PR-04963
9 pages
5 3 MEK 9100K SystemConnection
No ratings yet
5 3 MEK 9100K SystemConnection
68 pages
Teza de Abilitare COZMA Dragos LA 2019 Varianta Finala
No ratings yet
Teza de Abilitare COZMA Dragos LA 2019 Varianta Finala
189 pages
1 Introduction To Psychological Statistics
No ratings yet
1 Introduction To Psychological Statistics
83 pages
Applied Mathematics For Physical Chemistry
No ratings yet
Applied Mathematics For Physical Chemistry
117 pages
Phenomenological Theory of Low-Voltage Electroporation. Electric Field Calculations
No ratings yet
Phenomenological Theory of Low-Voltage Electroporation. Electric Field Calculations
9 pages
Domino Reactions in Organic Synthesis (Chem Rev 1996)
No ratings yet
Domino Reactions in Organic Synthesis (Chem Rev 1996)
22 pages
Herbal Radiomodulators Applications in Medicine, Homeland Defence and Space
No ratings yet
Herbal Radiomodulators Applications in Medicine, Homeland Defence and Space
352 pages
Curs de Proteomica Functionala Si Clinic
No ratings yet
Curs de Proteomica Functionala Si Clinic
424 pages
Webb-The Physics of Medical Imaging
No ratings yet
Webb-The Physics of Medical Imaging
318 pages
Ecg Viewer User Manual
No ratings yet
Ecg Viewer User Manual
34 pages
First Do No Harm
No ratings yet
First Do No Harm
88 pages
Si 1 Technology
No ratings yet
Si 1 Technology
85 pages
Medical Imaging: Sultana Alina Elena
No ratings yet
Medical Imaging: Sultana Alina Elena
28 pages
Orar Farmacie 2015 2016 S1 v2
100% (1)
Orar Farmacie 2015 2016 S1 v2
6 pages
Sante Dicom Viewer QSG
No ratings yet
Sante Dicom Viewer QSG
29 pages
A Comparative Analysis of Signal Processing and Classification Methods For Different Applications
No ratings yet
A Comparative Analysis of Signal Processing and Classification Methods For Different Applications
42 pages
Mol Dock
No ratings yet
Mol Dock
67 pages
Least Square Method
100% (2)
Least Square Method
14 pages
Fiziologie Aplicata - 2019
No ratings yet
Fiziologie Aplicata - 2019
16 pages
Advances in Quantitative Electroencephalogram Analysis Mathod
No ratings yet
Advances in Quantitative Electroencephalogram Analysis Mathod
46 pages
DSC (Differential Scanning Calorimetry)
No ratings yet
DSC (Differential Scanning Calorimetry)
30 pages
Univariate Time Series Modelling and Forecasting: Introductory Econometrics For Finance' © Chris Brooks 2002 1
No ratings yet
Univariate Time Series Modelling and Forecasting: Introductory Econometrics For Finance' © Chris Brooks 2002 1
62 pages
30 400-01 Vidas TSH: Summary and Explanation of The Test Principle of The Procedure
No ratings yet
30 400-01 Vidas TSH: Summary and Explanation of The Test Principle of The Procedure
9 pages
EEG Signal Processing For Medical Diagnosis Healthcare and Monitoring A Comprehensive Review
No ratings yet
EEG Signal Processing For Medical Diagnosis Healthcare and Monitoring A Comprehensive Review
27 pages
Chapter 9: Instrumentation On Bioreactor
No ratings yet
Chapter 9: Instrumentation On Bioreactor
34 pages
Sensors 23 06434 v2
No ratings yet
Sensors 23 06434 v2
27 pages
Problems Facing The Pharmaceutical Industry and Approaches To Ensure Long Term Viability
No ratings yet
Problems Facing The Pharmaceutical Industry and Approaches To Ensure Long Term Viability
68 pages
Indometacin
No ratings yet
Indometacin
47 pages
Advanced Applications On Antelope Database
No ratings yet
Advanced Applications On Antelope Database
29 pages
Signal Processing of EEG
No ratings yet
Signal Processing of EEG
27 pages
A Review On Machine Learning For EEG Signal Processing in Bioengineering
No ratings yet
A Review On Machine Learning For EEG Signal Processing in Bioengineering
15 pages
Histeroscopia Albu
No ratings yet
Histeroscopia Albu
29 pages
L1
No ratings yet
L1
2 pages
TEM Lecture 1
No ratings yet
TEM Lecture 1
33 pages
Biomechatronic Hand: Guide: Prof. P.Dhotrad Seminar By: Sagar.S.Kulkarni
No ratings yet
Biomechatronic Hand: Guide: Prof. P.Dhotrad Seminar By: Sagar.S.Kulkarni
16 pages
Guyton Part 2 PDF
100% (2)
Guyton Part 2 PDF
346 pages
FDA Draft Guidance - Diabetes Mellitus - Developing Drugs and Therapeutic Biologics For Treatment and Prevention
No ratings yet
FDA Draft Guidance - Diabetes Mellitus - Developing Drugs and Therapeutic Biologics For Treatment and Prevention
34 pages
Artificial Retina: BY Kusumanjali
No ratings yet
Artificial Retina: BY Kusumanjali
27 pages
Evaluation of Cellular Processes by in vitro Assays
From Everand
Evaluation of Cellular Processes by in vitro Assays
Taseen Gul
No ratings yet
1 PB
No ratings yet
1 PB
10 pages
PSUDO-CODE-Paper 107-A Novel Hybrid Deep Neural Network Classifier
No ratings yet
PSUDO-CODE-Paper 107-A Novel Hybrid Deep Neural Network Classifier
12 pages
Iso 9964 1 1993
No ratings yet
Iso 9964 1 1993
8 pages
10 1 s2.0 S1746809423000770 Main
No ratings yet
10 1 s2.0 S1746809423000770 Main
10 pages
NG Das Statistical Methods Volume 1
No ratings yet
NG Das Statistical Methods Volume 1
2 pages
ECOG Performance Status
No ratings yet
ECOG Performance Status
1 page
Electrophysiological Methods: Outline - Revision
No ratings yet
Electrophysiological Methods: Outline - Revision
5 pages
Bioseparation Citric Acid
No ratings yet
Bioseparation Citric Acid
32 pages
Hulley9781608318049 FM
67% (3)
Hulley9781608318049 FM
14 pages
Metode Spectrofotometrice C1
No ratings yet
Metode Spectrofotometrice C1
10 pages
Mass Spectrometry (MS) & Atomic Absorption Spectrometry (AAS)
No ratings yet
Mass Spectrometry (MS) & Atomic Absorption Spectrometry (AAS)
9 pages
Invelisul Celular: Plasmalema: - Separa Mediul Intracelular de Cel Extracelular
No ratings yet
Invelisul Celular: Plasmalema: - Separa Mediul Intracelular de Cel Extracelular
4 pages
(Oxford Science Publications) P.A. Cox - The Electronic Structure and Chemistry of Solids (1987, Clarendon Press)
100% (8)
(Oxford Science Publications) P.A. Cox - The Electronic Structure and Chemistry of Solids (1987, Clarendon Press)
134 pages
Bayesian Analysis Using MCMC On Survey Data
No ratings yet
Bayesian Analysis Using MCMC On Survey Data
35 pages
Forecasting
No ratings yet
Forecasting
73 pages
Metea Malina Stefania 432E Tema3
No ratings yet
Metea Malina Stefania 432E Tema3
2 pages
Niir Biopesticides Handbook
No ratings yet
Niir Biopesticides Handbook
5 pages
A Review On Analysis of EEG Signals PDF
No ratings yet
A Review On Analysis of EEG Signals PDF
4 pages
Capsule Gelatinoase
No ratings yet
Capsule Gelatinoase
19 pages
Physical Organic 1 Post-Lab
No ratings yet
Physical Organic 1 Post-Lab
7 pages
1694600777-Unit2.2 Logistic Regression CU 2.0
100% (1)
1694600777-Unit2.2 Logistic Regression CU 2.0
37 pages
Lab Report 1 (Handwritten)
No ratings yet
Lab Report 1 (Handwritten)
3 pages
Determinarea Cofeinei Prin HPLC
No ratings yet
Determinarea Cofeinei Prin HPLC
6 pages
Instrumentatie Virtuala - Subiecte de Examen - 2011/virtual Instrumentation - Exam Topics - 2011
No ratings yet
Instrumentatie Virtuala - Subiecte de Examen - 2011/virtual Instrumentation - Exam Topics - 2011
2 pages
Statistical Techniques in Business and Economics
No ratings yet
Statistical Techniques in Business and Economics
15 pages
Table of Areas Under The Normal Curve
No ratings yet
Table of Areas Under The Normal Curve
1 page
Lab 5 - Wireless Power Transfer PDF
No ratings yet
Lab 5 - Wireless Power Transfer PDF
2 pages
38 Metabolic Pathways 2
No ratings yet
38 Metabolic Pathways 2
1 page
Telecom Customer Churn Prediction Assessment-Pratik Zanke
No ratings yet
Telecom Customer Churn Prediction Assessment-Pratik Zanke
19 pages
Skewness, Five-Number Summary, Box-And-Whisker Plot and Kurtosis
No ratings yet
Skewness, Five-Number Summary, Box-And-Whisker Plot and Kurtosis
4 pages
Logistic Ordinal Regression
No ratings yet
Logistic Ordinal Regression
10 pages
Chapter 8 (Brooks) : Modelling Volatility and Correlation
No ratings yet
Chapter 8 (Brooks) : Modelling Volatility and Correlation
64 pages
Mgeb12 23S T21
No ratings yet
Mgeb12 23S T21
22 pages
Chapter - 4 Data Analysis and Interpretation
No ratings yet
Chapter - 4 Data Analysis and Interpretation
36 pages
The Levenberg Marquardt Algorithm For
No ratings yet
The Levenberg Marquardt Algorithm For
23 pages
Marketing Research: Data Analysis V: Regression Analysis (Part 1)
No ratings yet
Marketing Research: Data Analysis V: Regression Analysis (Part 1)
35 pages
Ch6 Multiple Regression
No ratings yet
Ch6 Multiple Regression
29 pages
examEC220 - IRDAP - Resit - Solutions
No ratings yet
examEC220 - IRDAP - Resit - Solutions
18 pages
Dougherty C12G02 2016 05 22
No ratings yet
Dougherty C12G02 2016 05 22
18 pages
Chi-Square Goodness of Fit Test and Its Applications
No ratings yet
Chi-Square Goodness of Fit Test and Its Applications
3 pages
3 XWEgynrp de 5 DPZ 4
No ratings yet
3 XWEgynrp de 5 DPZ 4
14 pages
Box Plots Questions MME
No ratings yet
Box Plots Questions MME
9 pages
Non-Inferiority Tests For Two Survival Curves Using Cox's Proportional Hazards Model
No ratings yet
Non-Inferiority Tests For Two Survival Curves Using Cox's Proportional Hazards Model
10 pages
Pengaruh Motivasi Terhadap Kinerja Karyawan Dengan Budaya Organisasi Sebagai Variabel Moderating (Studi Pada Pt. Randugarut Plastic Indonesia)
No ratings yet
Pengaruh Motivasi Terhadap Kinerja Karyawan Dengan Budaya Organisasi Sebagai Variabel Moderating (Studi Pada Pt. Randugarut Plastic Indonesia)
11 pages
Stat 200 Final Exam - The Latest Version - Umuc
0% (2)
Stat 200 Final Exam - The Latest Version - Umuc
3 pages
Cases Conjoint Analysis
No ratings yet
Cases Conjoint Analysis
5 pages
STA100 Fall2024 HW3 Due Oct 17th
No ratings yet
STA100 Fall2024 HW3 Due Oct 17th
2 pages
Notes 4 - Confidence Intervals and Significance Tests
No ratings yet
Notes 4 - Confidence Intervals and Significance Tests
1 page
Sathyabama University: Register Number
No ratings yet
Sathyabama University: Register Number
4 pages

A Review On Machine Learning For EEG Signal

Uploaded by

A Review On Machine Learning For EEG Signal

Uploaded by

This article has been accepted for publication in a future issue of this journal, but has not been

A Review on Machine Learning for EEG Signal

between variables. Various types of regression include: uni-

Author(s) Machine Learning Method Application Data Set Results

Pl Compared to other supervised and unsupervised learning

Author(s) Machine Learning Method Application Data Set Results

Fig. 8. A radial basis function network is an ANN which uses radial

state Ht. We can represent a simple Vanilla version of the

Author(s) Machine Learning Method Application Data Set Results

Author(s) Machine Learning Method Application Data Set Results

Author(s) Machine Learning Method Application Data Set Results

Author(s) Machine Learning Method Application Data Set Results

of truth for a given set of input variables. Fuzzy logic is

Author(s) Machine Learning Method Application Data Set Results

Author(s) Machine Learning Method Application Data Set

Author(s) Application Data Set Results

for emotion recognition, EEG is one of the most reliable

You might also like