基于代价敏感思想和自适应增强集成的SVM多分类算法

SVM Multi-classification Algorithm Based on Cost-sensitive Thought and AdaBoost Integration

针对数据识别分类在传统的支持向量机(SVM)个体分类器上正确识别率不理想的问题,提出一种基于代价敏感思想(cost-sensitive)和自适应增强(AdaBoost)的SVM集成数据分类算法(CAB-SVM)。在自适应增强算法每次迭代训练SVM弱分类器之前,根据样本总数设置初始样本权值,并抽取样本组成临时训练集训练SVM弱分类器。其中在权重迭代更新阶段,赋予被分错样本更高的误分代价,使得被分错样本权重增加更快,有效地减少了算法迭代次数。同时,算法迭代过程极大地优化了个体分类器的识别鲁棒性能,使得提出的CAB-SVM算法获得了更优越的数据分类性能。利用UCI数据样本集的实验结果表明CAB-SVM分类算法的正确识别率高于SVM和SVME算法。

Aiming at the problem that the correct recognition rate of data recognition and classification is not ideal on the traditional support vector machine(SVM)individual classifier,an integrated data classification algorithm based on cost-sensitive and AdaBoost SVM(CAB-SVM)is proposed.This method first sets the initial sample weights according to the total number of samples before training the SVM weak classifier in each iteration of AdaBoost algorithm,and extracts samples to form a temporary training set to train the SVM weak classifier.Among them,in the weight iteration update stage,a higher misclassification cost is given to the wrong sample,so that the weight of the wrong sample increases faster,and the number of algorithm iterations is effectively reduced.At the same time,the algorithm iterative process greatly optimizes the robust performance of individual classifiers,which makes the proposed CAB-SVM algorithm obtain better data classification performance.The experimental results,by the UCI data sample set,show that the correct recognition rate of the CAB-SVM classification algorithm is higher than that of the SVM and SVME algorithms.