ANAMOLY DETECTION TECHNIQUES USING SEQUENCES DATA

Abstract

Anomaly detection has traditionally dealt with record or transaction type data sets. But in many real domains, data naturally occurs as sequences, and there for e the desire of studying anomaly detection techniques in sequential data sets. The problem of detecting anomalies in sequence data sets is related to but differ ent from the traditional anomaly detection problem, becau se the nature of data and anomalies are differ ent than those found in record data sets. While th er e are many surveys and comparative evaluations for traditional anomaly detection, similar studies are not done for se quence anomaly detection. We in vestigate a broad spectrum of anomaly detec tion techniques for symbolic se quences, proposed in diverse app lication domains. Our hypothesis is that symbolic sequences from differ ent domains have distinct characteristics in terms of the nature of se quences as well as the nature of anomalies which makes it important to investigate how differ ent techniques behave for differ ent types of seque nce data. Such a study is criti cal to understand the relative strengths and weaknesses of differ ent techniques. Our paper is one such attempt where we have comparatively eva luated 7 anomaly detection tech niques on 10 public dat a sets, collected from thr ee diverse application domains. T o gain further understanding in the performance of the techniques, we present a novel way to generate sequence data with desired characteristics. The results on the artificiall y generated data sets help us in experimentally verifying our hypothesis regarding different techniques.