类不平衡模糊加权极限学习机算法研究

Research on Class Imbalance Fuzzy Weighted Extreme Learning Machine Algorithm

从理论上分析了样例不平衡分布对极限学习机性能产生危害的原因;在该理论框架下探讨了加权极限学习机在处理此类问题上的有效性及其固有缺陷;引入模糊集的思想,对传统的加权极限学习机进行了改进,并提出了4种用于解决类不平衡问题的模糊加权极限学习机算法;最后通过20个基准的二类不平衡数据集对所提算法的有效性和可行性进行了验证。实验结果表明:较之加权极限学习机及几种传统的不平衡极限学习机算法,提出的算法可明显获得更优的分类性能,并且与模糊加权支持向量机系列算法相比,所提算法通常可获得与之相当的分类性能,但时间开销往往更小。

Firstly,this paper analyzes the reason that the performance of extreme learning machine(ELM)is destroyedby imbalanced instance distribution in theory.Then,based on the same theoretical framework,this paper discussesthe effectiveness and inherent shortcomings of the weighted extreme learning machine(WELM).Nextly,profiting from the idea of fuzzy set,this paper proposes four fuzzy weighted extreme learning machine(FWELM)algorithms to deal with class imbalance problem.Finally,this paper verifies the effectiveness and feasibility of thesefour FWELM algorithms by the experiments constructing on20baseline binary-class imbalanced data sets.The experimentalresults indicate that the proposed algorithms can often acquire better classification performance than WELMalgorithm and several traditional class imbalance learning algorithms in the context of ELM.In addition,in contrast FSVMwithfuzzy support vector machine for class imbalance learning(FSVM-CIL)series algorithms,the proposed algorithmscan produce the comparable classification performance,but always consume less training time.