基于QPSO训练的SVM核函数集成学习研究

针对训练子集随机性强、规模大、算法时空复杂度高等问题,提出了基于量子微粒群的支持向量机（QPSO-SVM）核函数集成学习算法。该方法首先采用K-Means算法对训练样本进行聚类分析,然后根据其聚类分布选择少量具有代表性的样本,并通过基于量子行为的粒子群算法来训练单个支持向量机（SVM）,最后通过贝叶斯投票方法得到集成的SVM分类学习器。实验表明该方法在非线性高复杂度的数据分类中对分类精度有较大提高。

Study on Ensemble Learning for Kernel Selection Based on Quantum-behaved Particle Swarm Optimization Algorithm

Aiming at the existing problems in training subsets, which is strong randomicity,larger scale and high complexity. This paper proposes an ensemble learning approach for support vector machine （SVM） kernel selection based on QPSO （quantum-behaved particle swarm optimization algorithm）. Above all, the samples were clustered into several clusters using K-means analysis method. Then the small quantities of representative instances were chosen as training sets and with the samples to train SVM that adopt quantum-behaved particle swarm optimization algorithm to optimize the parameters. Ensemble improvement suppost vector machine classifier was constructed by Bayesian voting. The experimental results indicate that classification precision of this method has higher classification accuracy.