Scribd
Upload
12 views

Outlier Detection

Uploaded by

wasimrajaa
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPT, PDF, TXT or read online on Scribd
12 views

Outlier Detection

Uploaded by

wasimrajaa
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPT, PDF, TXT or read online on Scribd
You are on page 1/ 15

Outlier Discovery/Anomaly

Detection

Data Mining: Concepts and


December 10, 202 Techniques 1
4
Anomaly/Outlier Detection
 What are anomalies/outliers?
 The set of data points that are considerably

different than the remainder of the data


 Variants of Anomaly/Outlier Detection Problems
 Given a database D, find all the data points x  D

with anomaly scores greater than some threshold t


 Given a database D, find all the data points x  D

having the top-n largest anomaly scores f(x)


 Given a database D, containing mostly normal (but

unlabeled) data points, and a test point x, compute


the anomaly score of x with respect to D

Data Mining: Concepts and


December 10, 2024 Techniques 2
Applications

 Credit card fraud detection


 telecommunication fraud detection
 network intrusion detection
 fault detection
 many more

Data Mining: Concepts and


December 10, 2024 Techniques 3
Anomaly Detection
 Challenges
 How many outliers are there in the data?

 Method is unsupervised


Validation can be quite challenging (just like for
clustering)
 Finding needle in a haystack

 Working assumption:
 There are considerably more “normal”

observations than “abnormal” observations


(outliers/anomalies) in the data
Data Mining: Concepts and
December 10, 2024 Techniques 4
Anomaly Detection Schemes
 General Steps
 Build a profile of the “normal” behavior


Profile can be patterns or summary statistics for the
overall population
 Use the “normal” profile to detect anomalies

Anomalies are observations whose characteristics
differ significantly from the normal profile

 Types of anomaly detection


schemes
 Graphical & Statistical-based

 Distance-based

 Model-based

Data Mining: Concepts and


December 10, 2024 Techniques 5
Graphical Approaches
 Boxplot (1-D), Scatter plot (2-D), Spin plot
(3-D)

 Limitations
 Time consuming

 Subjective

Data Mining: Concepts and


December 10, 2024 Techniques 6
Convex Hull Method
 Extreme points are assumed to be outliers
 Use convex hull method to detect extreme
values

 What if the outlier occurs in the middle of the


data? Data Mining: Concepts and
December 10, 2024 Techniques 7
Statistical Approaches
 Assume a parametric model describing the distribution of the data
(e.g., normal distribution)

 Apply a statistical test that depends on


 Data distribution
 Parameter of distribution (e.g., mean, variance)
 Number of expected outliers (confidence limit)

Data Mining: Concepts and


December 10, 2024 Techniques 8
Grubbs’ Test
 Detect outliers in univariate data
 Assume data comes from normal distribution
 Detects one outlier at a time, remove the outlier,
and repeat
 H : There is no outlier in data
0

 HA: There is at least one outlier


 Grubbs’ test statistic: max X  X
G
s
 Reject H0 if:
( N  1) t (2 / N , N  2 )
G
N N  2  t (2 / N , N  2 )
Data Mining: Concepts and
December 10, 2024 Techniques 9
Statistical-based – Likelihood
Approach
 Assume the data set D contains samples from a mixture of
two probability distributions:
 M (majority distribution)
 A (anomalous distribution)
 General Approach:
 Initially, assume all the data points belong to M
 Let Lt(D) be the log likelihood of D at time t
 For each point xt that belongs to M, move it to A
 Let Lt+1 (D) be the new log likelihood.
 Compute the difference,  = Lt(D) – Lt+1 (D)
If  > c (some threshold), then xt is declared as an
anomaly and moved permanently from M to A

Data Mining: Concepts and


December 10, 2024 Techniques 10
Limitations of Statistical
Approaches
 Most of the tests are for a single attribute

 In many cases, data distribution may not be


known

 For multi-dimensional data, it may be


difficult to estimate the true distribution

Data Mining: Concepts and


December 10, 2024 Techniques 11
Distance-based Approaches
 Data is represented as a vector of features

 Three major approaches


 Nearest-neighbor based

 Density based

 Clustering based

Data Mining: Concepts and


December 10, 2024 Techniques 12
Nearest-Neighbor Based
Approach
 Approach:
 Compute the distance between every pair of data points

 There are various ways to define outliers:



Data points for which there are fewer than p
neighboring points within a distance D


The top n data points whose distance to the kth
nearest neighbor is greatest


The top n data points whose average distance to the k
nearest neighbors is greatest

Data Mining: Concepts and


December 10, 2024 Techniques 13
Density-based: LOF approach
 For each point, compute the density of its local
neighborhood
 Compute local outlier factor (LOF) of a sample p as the
average of the ratios of the density of sample p and the
density of its nearest neighbors
 Outliers are points with largest LOF value

In the NN approach, p2
is not considered as
outlier, while LOF
approach find both p1
p2 and p2 as outliers
 p1

Data Mining: Concepts and


December 10, 2024 Techniques 14
Clustering-Based
 Basic idea:
 Cluster the data into groups

of different density
 Choose points in small

cluster as candidate
outliers
 Compute the distance

between candidate points


and non-candidate clusters.

If candidate points are
far from all other non-
candidate points, they
are outliers

Data Mining: Concepts and


December 10, 2024 Techniques 15

You might also like