Incremental Weighted and Adaptive Sliding Window Based Classification Algorithm for Imbalanced Data Stream

Concept drift and class imbalance are significant challenges in data stream learning, impacting algorithmic classification performance. This study introduces an online updating mechanism to a data-block ensemble method, incorporating resampling and a forgetting mechanism alongside adaptive sliding window technology. We propose Adaptive Window and Incremental Weighting for Imbalance Learning (AWIWIL), an innovative method that operates on an ensemble framework. AWIWIL utilizes a variable-size window forgetting mechanism to assess the recent performance of base classifiers and dynamically adjusts their weights as new instances arrive. Experimental evaluations on both synthetic and real datasets demonstrate that AWIWIL effectively adapts to concept drift and excels in G-mean and Recall metrics.