一种鲁棒非平衡极速学习机算法  

Robust extreme learning machine algorithm based on imbalanced datasets

作　　者：孟凡荣[1] 高春晓[1] 刘兵[1] 

机构地区：[1]中国矿业大学计算机科学与技术学院,江苏徐州221116

出　　处：《计算机应用研究》2014年第4期985-988,1004,共5页Application Research of Computers

基　　金：高等学校博士学科点专项科研基金资助项目(20110095110010);国家"863"计划资助项目(2012AA011004)

摘　　要：极速学习机(ELM)算法只对平衡数据集分类较好,对于非平衡数据集,它通常偏向多数样本类,对于少数样本类性能较低。针对这一问题,提出了一种处理不平衡数据集分类的ELM模型(ELM-CIL),该模型按照代价敏感学习的原则为少数类样本赋予较大的惩罚系数,并引入模糊隶属度值减小了外围噪声点的影响。实验表明,提出的方法不仅对提高不平衡数据集中少数类的分类精度效果较明显,而且提高了对噪声的鲁棒性。ELM algorithm can achieve better classification results for balanced datasets. For the imbalanced datasets, it is u- sually favor of the majority class and has lower class performance for a small number of samples. To solve this problem, this paper proposed an ELM model （ELM-CIL model） dealing with imbalanced dataset classification problems. The model was in accordance with the principles of cost-sensitive learning fhat gave a greater penalty coefficient for the minority class sample. At the same time, the introduction of fuzzy membership value reduced the impact of external noise points. The experiments show that the proposed method is more obvious to improve minority class classification accuracy and has better robustness to noise than traditional ELM algorithm.

关 键 词：极速学习机 不平衡数据集 基于核的可能性模糊C-均值聚类 神经网络 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]