Diversity Measure for Concept Drift Detection in Data Streams

Concept drift is a notable challenge in machine learning, data mining, and applications involving big data and large-scale data processing. The employment of diversity measures has emerged as an effective strategy. We examine and investigate the role of diversity measures in detecting concept drift and provide a comparative analysis of four different approaches: DMDDM for drift detection in a fully supervised binary classification context, DMDDM-S in a semi-supervised context, DMODD for online drift detection in a fully supervised multi-classification context, and HBBE, a hybrid block-based ensemble designed for addressing different types of concept drifts. Our comparative analysis evaluates the efficacy of these methods in detecting concept drift and enhancing model performance. The results confirm the effectiveness of all four approaches within their respective settings. Moreover, this paper provides insights into potential advancements and research opportunities in the application of diversity measures for concept drift detection.