基于粒子群算法的决策树SVM多分类方法研究

Study on PSO-based decision-tree SVM multi-class classification method

针对SVM多分类问题提出了一种基于粒子群算法的最优决策树SVM生成算法,以解决传统支持向量机多分类方法存在的不可分区域和误差积累现象。该方法利用自变异的PSO聚类算法在每一决策节点自动寻找最优或近优分类决策,将数据集划分为两类,直至叶子节点为止,最终根据最优决策树构建SVM多分类结构,训练各个节点SVM分类器。将该算法应用于图像人群密度分类问题,仿真实验表明,分类精度和分类时间得到明显改善,是一种有效地的多分类算法。

This paper proposes a PSO-based decision tree SVM multi-class classification algorithm to resolve unclassi-fiable region and error accumulation phenomenon existing in traditional support vector machine multi-class classifica-tion method.PSO algorithm is used to cluster the dataset into two patterns on each node, search optimal or subopti-mal decision-tree automatically, and then construct the SVM classifier with the optimal decision tree.The algorithm is applied to the crowd density image classification problem .The simulation results show that classification accuracy and time are improved obviously, and it is an effective multi-class classification algorithm.