Evolutionary-Based Weighted Broad Learning System for Imbalanced Learning
Tianlun Zhang, Yang Li, Rong Chen · 2019
As a novel family of connectionist model, broad learning system (BLS) with the form of flat structure is established for broadly-fused feature representation. As a result of two assumptions, namely, the same misclassification cost and balanced class distribution, BLS usually fails to achieve favorable classification performance in the presence of severely skewed category distribution. To deal with this problem, we develop BLS as a cost-sensitive classification learning that assigns different misclassification costs for different categories, and minimizes the cumulative classification cost. To be more actionable, we formulate the setting of penalty factors as a multi-objective optimization problem, which can be solved by an evaluation algorithms. Thus, without rules of thumb, the proposed BLS can address various imbalanced tasks in an adaptive manner. Moreover, due to the sparse form of the cost matrix we design, the optimization problem has less objectives to be solved. Comprehensive experiments have been conducted on several benchmark and real-world data sets, the experimental results show that the proposed method achieves competitive results compared with the existing state-of-the-art imbalanced learning.