Data Classification

Chapter 17

Rare Class Learning

Charu C. Aggarwal

IBM T. J. Watson Research CenterYorktown Heights, NY 10598 [e-mail]

17.1 Introduction

The problem of rare class detection is closely related to outlier analysis [2]. In unsupervised outlier analysis, no supervision is used for the anomaly detection process. In such scenarios, many of the anomalies found correspond to noise, and may not be of any interest to an analyst. It has been observed [35, 42, 61] in diverse applications such as system anomaly detection, financial fraud, and Web robot detection that the nature of the anomalies is often highly specific to particular kinds of abnormal activity in the underlying application. In such cases, unsupervised outlier detection methods may often ...