Scribd
Upload
212 views

Outlier Detection

The document discusses outlier detection and anomaly detection in data mining. It defines outliers as data points that are considerably different from the majority of data points. It describes different types of anomaly detection problems and applications. It also outlines several approaches to anomaly detection, including graphical, statistical, distance-based, and clustering-based methods. Each approach has its own advantages and limitations for detecting outliers in datasets.

Uploaded by

Savitha Vasanthan
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
212 views

Outlier Detection

The document discusses outlier detection and anomaly detection in data mining. It defines outliers as data points that are considerably different from the majority of data points. It describes different types of anomaly detection problems and applications. It also outlines several approaches to anomaly detection, including graphical, statistical, distance-based, and clustering-based methods. Each approach has its own advantages and limitations for detecting outliers in datasets.

Uploaded by

Savitha Vasanthan
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/ 19

Outlier Discovery/Anomaly Detection

Data Mining: Concepts and


July 12, 2019 Techniques 1
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

July 12, 2019 Data Mining: Concepts and Techniques 2


Applications

 Credit card fraud detection


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

July 12, 2019 Data Mining: Concepts and 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
July 12, 2019 Data Mining: Concepts and 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

July 12, 2019 Data Mining: Concepts and Techniques 5


Graphical Approaches

 Boxplot (1-D), Scatter plot (2-D), Spin plot (3-D)

 Limitations
 Time consuming

 Subjective

July 12, 2019 Data Mining: Concepts and 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?
July 12, 2019 Data Mining: Concepts and 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)

July 12, 2019 Data Mining: Concepts and 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
 H0: There is no outlier in data

 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 )
July 12, 2019 Data Mining: Concepts and 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

July 12, 2019 Data Mining: Concepts and Techniques 10


Statistical-based – Likelihood
Approach
 Data distribution, D = (1 – ) M +  A
 M is a probability distribution estimated from data
 Can be based on any modeling method

 A is initially assumed to be uniform distribution

 Likelihood at time t:
N   |At | 
Lt ( D )   PD ( xi )   (1   )  PM t ( xi )    PAt ( xi ) 
|M t |

i 1  xi M t  xiAt 
LLt ( D )  M t log( 1   )   log PM t ( xi )  At log    log PAt ( xi )
xi M t xi At

July 12, 2019 Data Mining: Concepts and Techniques 11


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

July 12, 2019 Data Mining: Concepts and Techniques 12


Distance-based Approaches

 Data is represented as a vector of features

 Three major approaches


 Nearest-neighbor based

 Density based

 Clustering based

July 12, 2019 Data Mining: Concepts and Techniques 13


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

July 12, 2019 Data Mining: Concepts and Techniques 14


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 and p2 as outliers
p2
 p1

July 12, 2019 Data Mining: Concepts and Techniques 15


LOF

The local outlier factor LOF, is defined as follows:


lrd k (o)
oNk ( p ) lrd ( p)
LOFk ( p)  k
| N k ( p) |
where Nk(p) is the set of k-nearest neighbors to p
| N k ( p) |
and lrd k ( p) 
 oN k ( p)
reach  dist ( p, o)

reach  dist k ( p)  max{ k  dist (o), dist ( p, o)}

July 12, 2019 Data Mining: Concepts and Techniques 16


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

July 12, 2019 Data Mining: Concepts and Techniques 17


Outliers in Lower Dimensional Projection

 Divide each attribute into  equal-depth intervals


 Each interval contains a fraction f = 1/ of the records

 Consider a d-dimensional cube created by picking grid ranges


from d different dimensions
 If attributes are independent, we expect region to contain
a fraction fk of the records
 If there are N points, we can measure sparsity of a cube
D as:

 Negative sparsity indicates cube contains smaller number


of points than expected
 To detect the sparse cells, you have to consider all cells….
exponential to d. Heuristics can be used to find them…
July 12, 2019 Data Mining: Concepts and Techniques 19
Example

 N=100,  = 5, f = 1/5 = 0.2, N  f2 = 4

July 12, 2019 Data Mining: Concepts and Techniques 20

You might also like