关键词: 扰动, 粗糙集, 相对核, 动态约简, 负相关学习, 加速遗传算法, 支持向量机集成

Abstract: This paper proposed a selective Support Vector Machine (SVM) ensemble algorithm based on double disturbance to improve the generalization ability of SVM. First, the training samples were disturbed by using conventional boosting algorithm, a dynamic reduction algorithm, which integrated relative reduction based on relative core of rough set and resample method, to produce individual SVM. The fitness function of genetic factors was established based on negative correlation learning, and Best SVM with weight larger than a given threshold value were selected by accelerating genetic algorithm and were integrated using weighted average. The experiments show that the algorithm has higher generalization performance, and lower time and space complexity. It is a highly effective ensemble algorithm.

Key words: disturbance, rough set, relative core, dynamic reduction, negative correlation learning, Accelerating Genetic Algorithm (AGA), support vector machine ensemble