论文部分内容阅读
针对传统分类算法隐含的假设(相信并且接受每个样本的分类结果)在医疗/故障诊断和欺诈/入侵检测等领域中并不适用的问题,提出嵌入非对称拒识代价的二元分类问题,并对其进行简化.在此基础上设计出基于支持向量机(SVM)的代价敏感分类算法(CSVM-CRC).该算法包括训练SVM分类器、计算后验概率、估计分类可靠性和确定最优拒识阈值4个步骤.基于10个Benchmark数据集的实验研究表明,CSVM-CRC算法能够有效降低平均代价.
Aiming at the problem that traditional classification algorithm implicit assumption (believe and accept the classification result of each sample) does not apply in medical / fault diagnosis and fraud / intrusion detection, this paper proposes a binary classification problem with asymmetric rejection cost (CSVM-CRC) based on Support Vector Machine (SVM) is proposed in this paper.The algorithm includes training SVM classifier, calculating posterior probability, estimating classification reliability and determining Optimal rejection threshold of 4. The experimental study based on 10 Benchmark datasets shows that the CSVM-CRC algorithm can effectively reduce the average cost.