Quasi-linear SVM classifier with segmented local offsets for imbalanced data classification

Within-class imbalance problems often occur in imbalanced data classification, which worsen the imbalance distribution problem and increase the learning concept complexity. However, most existing methods for the imbalanced data classification focus on rectifying the between-class imbalance problem, which is insufficient and inappropriate in many different scenarios. This paper proposes a simple yet effective support vector machine (SVM) classifier with local offset adjustment for imbalance classification problems. First, a geometry-based partitioning method is modified for imbalanced datasets to divide the input space into multiple linearly separable partitions along the potential separation boundary. Then an F-score-based method is applied to obtain local offsets optimized on each local cluster. Finally, by constructing a quasi-linear kernel based on the partitioning information, a quasi-linear SVM classifier with local offsets is constructed for the imbalanced datasets. Simulation results on different real-world datasets show that the proposed method is effective for imbalanced data classifications.