关键词: 类不平衡, 极限学习机, 代价敏感学习, 样本先验信息

Abstract: Extreme learning machine can be widely used in classification, clustering, regression, etc. However, previous researchers ignore the influence of sample prior distribution information for classification performance when they deal with class imbalance problems. Aiming at this problem, this paper presents an algorithm called CPWELM ( Coupling sample Prior distribution Weighted Extreme Learning Machine), which is based on extreme learning machine. We fully discuss the importance of the different distribution sample points, then we construct the cost matrix with it for the improvement of classifier performance. We do experiments on 12 imbalanced datasets to verify the feasibility and effectiveness of the proposed algorithm. The results indicate that the proposed algorithm generally performs better than the state-of-the-art ones.

Key words: class imbalance, extreme learning machine, cost-sensitive learning, sample prior information