CAJ | 학술논문

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。
침대SVM다분류문제제출료일충기우입자군산법적최우결책수SVM생성산법，이해결전통지지향량궤다분류방법존재적불가분구역화오차적루현상。해방법이용자변이적PSO취류산법재매일결책절점자동심조최우혹근우분류결책，장수거집화분위량류，직지협자절점위지，최종근거최우결책수구건SVM다분류결구，훈련각개절점SVM분류기。장해산법응용우도상인군밀도분류문제，방진실험표명，분류정도화분류시간득도명현개선，시일충유효지적다분류산법。
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.