An improved under sampling approaches for concept drift and class imblance data streams using PSO and BEE search algorithms

Abstract

Data Streams are one of the research areas in knowledge discovery from data sources of advance equipments generating huge volumes of data. In data streams two unique dimensions of complexities arise due to Concept Drift and Class imbalance handling. Two solve both the issues simultaneously, in this paper we have proposed a Concept Drift with Class Imbalance Handling Approach using two nature inspired evolutionary searching techniques known as Evolutionary Under Sampling using Particle Swarm Optimization (EUPSO) and Evolutionary Under Sampling using Bee Search (EUBS) algorithm. The key rationale behind the new proposals is to apply optimized search to deal simultaneously both concept drift with class imbalance. The bridging of two separately developed areas of concept drift and class imbalance in data stream learning is well mapped to converge the proposals under unique conditions. The experimental results and evidences suggest that the proposed approaches perform better than the existing ones.