Recurring and novel class detection in concept-drifting data streams using class-based ensemble

Abstract

Over the recent years concept-evolution has received a lot of attention because of its importance in the context of mining data streams. Mining data stream has become an important task due to its wide range of applications such as network intrusion detection, credit card fraud protection, identifying trends in the social networks etc. Concept-evolution means introduction of novel class in the data stream. Many recent works address this phenomenon. In addition, a class may appear in the stream, disappears for a while and then reemerges. This scenario is known as recurring classes and remained unaddressed in most of the cases. As a result, generally where a novel class detection system is present, any recurring class is falsely detected as novel class. This results in unnecessary waste of human and computational resources. In this paper, we have proposed a class-based ensemble of classification model addressing the issues of recurring and novel class in the presence of concept drift and noise. Our approach has shown impressive performance compared to the state-of-art methods in the literature. © 2014 Springer International Publishing.