Outlier DiscoveryAnomaly Detection 6102021 Data Mining Concepts and

Outlier Discovery/Anomaly Detection 6/10/2021 Data Mining: Concepts and Techniques 1

Anomaly/Outlier Detection n n What are anomalies/outliers? n The set of data points that are considerably different than the remainder of the data Variants of Anomaly/Outlier Detection Problems n Given a database D, find all the data points x D with anomaly scores greater than some threshold t n Given a database D, find all the data points x D having the top-n largest anomaly scores f(x) n 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 6/10/2021 Data Mining: Concepts and Techniques 2

Applications n n n 6/10/2021 Credit card fraud detection telecommunication fraud detection network intrusion detection fault detection many more Data Mining: Concepts and Techniques 3

Anomaly Detection n Challenges n How many outliers are there in the data? n Method is unsupervised Validation can be quite challenging (just like for clustering) n n n Finding needle in a haystack Working assumption: n There are considerably more “normal” observations than “abnormal” observations (outliers/anomalies) in the data 6/10/2021 Data Mining: Concepts and Techniques 4

Anomaly Detection Schemes n General Steps n Build a profile of the “normal” behavior n n Use the “normal” profile to detect anomalies n n Profile can be patterns or summary statistics for the overall population Anomalies are observations whose characteristics differ significantly from the normal profile Types of anomaly detection schemes n Graphical & Statistical-based n Distance-based n Model-based 6/10/2021 Data Mining: Concepts and Techniques 5

Graphical Approaches n n Boxplot (1 -D), Scatter plot (2 -D), Spin plot (3 -D) Limitations n Time consuming n Subjective 6/10/2021 Data Mining: Concepts and Techniques 6

Convex Hull Method n n n 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? 6/10/2021 Data Mining: Concepts and Techniques 7

Statistical Approaches n n Assume a parametric model describing the distribution of the data (e. g. , normal distribution) Apply a statistical test that depends on n Data distribution n Parameter of distribution (e. g. , mean, variance) n Number of expected outliers (confidence limit) 6/10/2021 Data Mining: Concepts and Techniques 8

Grubbs’ Test n Detect outliers in univariate data Assume data comes from normal distribution Detects one outlier at a time, remove the outlier, and repeat n H 0: There is no outlier in data n HA: There is at least one outlier Grubbs’ test statistic: n Reject H 0 if: n n n 6/10/2021 Data Mining: Concepts and Techniques 9

Statistical-based – Likelihood Approach n n Assume the data set D contains samples from a mixture of two probability distributions: n M (majority distribution) n A (anomalous distribution) General Approach: n Initially, assume all the data points belong to M n Let Lt(D) be the log likelihood of D at time t n For each point xt that belongs to M, move it to A n Let Lt+1 (D) be the new log likelihood. n Compute the difference, = Lt(D) – Lt+1 (D) n If > c (some threshold), then xt is declared as an anomaly and moved permanently from M to A 6/10/2021 Data Mining: Concepts and Techniques 10

Statistical-based – Likelihood Approach n n n Data distribution, D = (1 – ) M + A M is a probability distribution estimated from data n Can be based on any modeling method n A is initially assumed to be uniform distribution Likelihood at time t: 6/10/2021 Data Mining: Concepts and Techniques 11

Limitations of Statistical Approaches n n n 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 6/10/2021 Data Mining: Concepts and Techniques 12

Distance-based Approaches n n Data is represented as a vector of features Three major approaches n Nearest-neighbor based n Density based n Clustering based 6/10/2021 Data Mining: Concepts and Techniques 13

Nearest-Neighbor Based Approach n Approach: n Compute the distance between every pair of data points n There are various ways to define outliers: n Data points for which there are fewer than p neighboring points within a distance D n n 6/10/2021 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 Techniques 14

Density-based: LOF approach n n n 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, p 2 is not considered as outlier, while LOF approach find both p 1 and p 2 as outliers p 2 6/10/2021 p 1 Data Mining: Concepts and Techniques 15

LOF The local outlier factor LOF, is defined as follows: where Nk(p) is the set of k-nearest neighbors to p and 6/10/2021 Data Mining: Concepts and Techniques 16

Clustering-Based n Basic idea: n Cluster the data into groups of different density n Choose points in small cluster as candidate outliers n Compute the distance between candidate points and noncandidate clusters. n If candidate points are far from all other non-candidate points, they are outliers 6/10/2021 Data Mining: Concepts and Techniques 17

Outliers in Lower Dimensional Projection n n Divide each attribute into equal-depth intervals n Each interval contains a fraction f = 1/ of the records Consider a d-dimensional cube created by picking grid ranges from d different dimensions n If attributes are independent, we expect region to contain a fraction fk of the records n If there are N points, we can measure sparsity of a cube D as: n n 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… 6/10/2021 Data Mining: Concepts and Techniques 19

Example n N=100, = 5, f = 1/5 = 0. 2, N f 2 = 4 6/10/2021 Data Mining: Concepts and Techniques 20