Novel and Recurring Class Detection using Ensemble of Classifiers: A Class-based Approach

Over the recent years, concept-evolution has received a lot of attention to the research community because of its importance in the context of mining data streams. Mining data stream has become a crucial task due to its wide range of applications such as network intrusion detection, credit card fraud identification, 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 also 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 investigated the idea of a class-based ensemble of classification model addressing the issues of recurring and novel class in the presence of concept drift. Our approach has shown impressive performance compared to the state-of-art methods in the literature.