基于加速遗传算法的选择性支持向量机集成

摘要

为有效提升支持向量机的泛化性能,提出基于加速遗传算法的选择性支持向量机集成.通过Bootstrap技术产生并训练得到多个独立子SVM,基于负相关学习理论构造适应度函数,提高子SVM的泛化性能,并增大其之间差异度.利用加速遗传算法计算各子SVM在加权平均中的最优权重,然后选择权值大于一定阈值的部分SVM进行加权集成.实验结果表明,该算法是一种有效的集成方法,能进一步提高SVM的集成效率和泛化性能.%This paper presented selective SVM ensemble based on accelerating genetic algorithm to improve the generalization ability of SVM.Produced many SVM by Bootstrap methods, established the fitness function based on negative correlation learning to improve generalization and high dissimilarity with others.Calculated the weighte of SVM by accelerating genetic algorithm, then ensembled those SVMs with weight larger than a given threshold value using weights average.Experiments results show that the algorithm is an effect ensemble method and improves the ensemble efficiency and generalization ability of SVM.