1

COMPARISON OF ML ALGORITHMS
VARIOUS ML ALGORITHMS

PROS & CONS

Algorithm Best at Pros Cons
Random Forest Apt at almost any Can work in parallel Difficult to interpret
machine learning
problem Seldom overfits Weaker on regression when
estimating values at the
Bioinformatics Automatically handles extremities of the distribution
missing values of response values

No need to transform any Biased in multiclass problems

variable toward more frequent classes

No need to tweak
parameters

Can be used by almost

anyone with excellent
results
Linear regression Baseline predictions Simple to understand and You have to work hard to make
explain it fit nonlinear functions
Econometric
predictions It seldom overfits Can suffer from outliers

Modelling marketing Using L1 & L2

responses regularization is effective
in feature selection

Fast to train

Easy to train on big data

thanks to its stochastic
version
Support Vector Character recognition Automatic nonlinear Difficult to interpret when
Machines feature creation applying nonlinear kernels
Image recognition
Can approximate complex Suffers from too many
Text classification nonlinear functions examples, after 10,000
examples it starts taking too
long to train
K-nearest Computer vision Fast, lazy training Slow and cumbersome in the
Neighbors predicting phase
Multilabel tagging Can naturally handle

BUSINESS ANALYTICS
 2

extreme multiclass
Recommender problems (like tagging Can fail to predict correctly due
systems text) to the curse of dimensionality

Spell checking
problems
Naive Bayes Face recognition Easy and fast to Strong and unrealistic feature
implement, doesn’t independence assumptions
Sentiment analysis require too much memory
and can be used for online Fails estimating rare
Spam detection learning occurrences

Text classification Easy to understand Suffers from irrelevant features

Takes into account prior

knowledge
Logistic regression Ordering results by Simple to understand and You have to work hard to make
probability explain it fit nonlinear functions

Modelling marketing It seldom overfits Can suffer from outliers

responses
Using L1 & L2
regularization is effective
in feature selection

The best algorithm for

predicting probabilities of
an event

Fast to train

Easy to train on big data

thanks to its stochastic
version
K-means Segmentation Fast in finding clusters Suffers from multicollinearity

Can detect outliers in Clusters are spherical, can’t

multiple dimensions detect groups of other shape

Unstable solutions, depends on

initialization

BUSINESS ANALYTICS
 3

CALCULATIONS AND INTERPRETATION OF

RESULTS
NAÏVE BAYES

NAÏVE BAYES ALGORITHM: THEORY AND CALCULATIONS

Bayes theorem provides a way of calculating the posterior probability, P(c|x), from P(c), P(x), and P(x|c). Naive Bayes
classifier assume that the effect of the value of a predictor (x) on a given class (c) is independent of the values of
other predictors. This assumption is called class conditional independence.

 P(c|x) is the posterior probability of class (target) given predictor (attribute).

 P(c) is the prior probability of class.
 P(x|c) is the likelihood which is the probability of predictor given class.
 P(x) is the prior probability of predictor.

The posterior probability can be calculated by first, constructing a frequency table for each attribute against the
target. Then, transforming the frequency tables to likelihood tables and finally use the Naive Bayesian equation to
calculate the posterior probability for each class. The class with the highest posterior probability is the outcome of
prediction.

BUSINESS ANALYTICS
 4

Joint posterior probability for 8 independent variables is a logical extension of two variables, by taking product of
all 8 conditional probabilities. Since python library or any statistical software would do the math, only
demonstration of logic and a sample calculation is adequate.

PRESENTATION OF RESULTS: CONFUSION MATRIX

Learners are expected to build tables of likelihood for 8 variables. (Depending on time available they may do
so only for one or two variables.)

 True positives (TP): These are cases in which we predicted yes (they have the disease),
and they do have the disease.
 True negatives (TN): We predicted no, and they don't have the disease.
 False positives (FP): We predicted yes, but they don't actually have the disease. (Also
known as a "Type I error.")
 False negatives (FN): We predicted no, but they actually do have the disease. (Also
known as a "Type II error.")

Sample Confusion matrix (exact numbers to be populated from Diabetes case)

BUSINESS ANALYTICS
 5

INTERPRETATION OF RE SULTS: TESTING ACCURACY

Accuracy is not a single number but depending on precise question being answered, can be from amongst :-

 Accuracy: Overall, how often is the classifier correct?

o (TP+TN)/total = (100+50)/165 = 0.91
 Misclassification Rate: Overall, how often is it wrong?
o (FP+FN)/total = (10+5)/165 = 0.09
o equivalent to 1 minus Accuracy
o also known as "Error Rate"
 True Positive Rate: When it's actually yes, how often does it predict yes?
o TP/actual yes = 100/105 = 0.95
o also known as "Sensitivity" or "Recall"
 False Positive Rate: When it's actually no, how often does it predict yes?
o FP/actual no = 10/60 = 0.17
 Specificity: When it's actually no, how often does it predict no?
o TN/actual no = 50/60 = 0.83
o equivalent to 1 minus False Positive Rate
 Precision: When it predicts yes, how often is it correct?
o TP/predicted yes = 100/110 = 0.91
 Prevalence: How often does the yes condition actually occur in our sample?
o actual yes/total = 105/165 = 0.64

SUPPORT VECTOR MACHINES

THEORY AND CALCULATIONS

The algorithm is best explained using visualization and basic concepts in coordinate geometry.

BUSINESS ANALYTICS
 6

INTERPRETATION OF WE IGHTS

Magnitude of weight after due normalization of variables, helps in deciding feature importance. The interpretation
of weights can be explained using OLS (Ordinary Least Squares) regression as an analogy.

BUSINESS ANALYTICS
 7

Detailed working example can be used from :-

https://charlienewey.github.io/manually-calculating-an-svms-support-vectors/

INTERPRETATION OF RE SULTS: TESTING ACCURACY

While exact values of results will vary between Naïve Bayes and SVM, the interpretation of testing accuracy is
same. Previous section on Naïve Bayes accuracy refers.

BUSINESS ANALYTICS
 8

HEALTHCARE APPLICATIONS
USING ML ALGORITHMS

The value of machine learning in healthcare is its ability to process huge datasets beyond the scope of human
capability, and then reliably convert analysis of that data into clinical insights that aid physicians in planning and
providing care, ultimately leading to better outcomes, lower costs of care, and increased patient satisfaction.

It has been estimated that big data and machine learning in pharma and medicine could generate a value of up to
$100B annually, based on better decision-making, improved efficiency of research/clinical trials, and new tool
creation for physicians, consumers, insurers, and regulators.

a. Disease identification / diagnosis - Creation of a platform to analyze data, and loop it back in real time to
physicians to aid in clinical decision making is CDSS. A physician sees a patient and enters symptoms, data,
and test results into the EMR, there’s machine learning behind the scenes looking at everything about that
patient, and prompting the doctor with useful information for making a diagnosis, ordering a test, or
suggesting a preventive screening. In the long term, we will be able to incorporate bigger sets of data that
can be analyzed in real time to provide all kinds of information to the provider and patient.
b. Show causal relationships in disease prognosis and help in predictions.
c. Patient risk profile – depending on various signs and symptoms and lifestyle factors.
d. Gather public health data and predict epidemic outbreaks.
e. Reduce 1-year mortality - Health systems can reduce 1-year mortality rates by predicting the likelihood of
death within one year of discharge and then match patients with appropriate interventions, care providers,
and support.

NEW SERVICES & PRODUC TS- CLINICAL

1. Application of ML classification algorithms on diabetes dataset is representative case study of
classifying any clinical dataset into infected/not-infected categories. For example, likelihood of
cancer, hypertension etc, through careful selection of variables relevant to predicting that
particular disease.
2. Predict chronic disease - Machine learning can help hospital systems identify patients with
undiagnosed or misdiagnosed chronic disease, predict the likelihood that patients will develop
chronic disease, and present patient-specific prevention interventions.
3. If time-series data is available then diseases can be predicted based on ML from historic data.
4. New products can be developed by integrating ML algorithms into existing diagnostic solutions.
For example, medical image classification uses similar ML classifiers to classify retinal images into
diabetic retinopathy absent/present. Such diagnostic reports can be made more informative for
radiologists by providing ML-provided insights.

NEW SERVICES & PRODUC TS- HEALTHCARE MANAGEMENT

a. Reduce readmissions - Machine learning can reduce readmissions in a targeted, efficient, and patient-
centered manner. Clinicians can receive daily guidance as to which patients are most likely to be readmitted
and how they might be able to reduce that risk.
b. Prevent hospital acquired infections (HAIs). Clinicians can monitor high risk patients and intervene to
reduce that risk by focusing on patient-specific risk factors.

BUSINESS ANALYTICS
 9

c. Reduce hospital Length-of-Stay (LOS). Health systems can reduce LOS and improve other outcomes like
patient satisfaction by identifying patients that are likely to have an increased LOS and then ensure that
best practices are followed.
d. Predict propensity-to-pay - Health systems can determine who needs reminders, who needs financial
assistance, and how the likelihood of payment changes over time and after particular events.
e. Predict no-shows - Health systems can create accurate predictive models to assess, with each scheduled
appointment, the risk of a no-show, ultimately improving patient care and the efficient use of resources.

BUSINESS ANALYTICS