Biclustering-KNN joint learning in anomaly detection for handling class-imbalance-problem

Classification of imbalanced data sets is a major challenge in machine learning. To overcome the limitations of existing methods, this paper proposes an ensembled bicluster-based classification method (EBC) approach to address the problem of imbalanced data. This method combines biclustering algorithms, anomaly detection algorithms, and ensemble learning to perform classification tasks on imbalanced datasets. Existing methods model majority samples in specific feature spaces, leading to learning relatively limited distributions. EBC integrates bicluster with anomaly detection to construct basic classifiers, allowing the model to learn more diverse distributions. Additionally, many existing methods only model majority samples, which may result in misclassifications. EBC adopts a two-stage classification framework, where the first stage models majority samples and the second stage models minority samples. This approach not only improves the classification accuracy of minority samples but also ensures the accuracy of majority samples, ultimately enhancing the model's generalization performance. To ensure diversity among basic classifiers, each basic classifier in EBC corresponds to a unique biclustering result comprising different samples and features. Comparative experiments between the proposed algorithm and several typical algorithms demonstrate the effectiveness of the proposed algorithm. It performed best in 20 groups of comparative experiments with different imbalance ratios. The average GMeans on all data sets is 85.96%, which is 22.27% better than baseline, significantly ahead of other comparison algorithms.