基于SVM-RFE特征选择的规则提取方法

Rule Extraction Method Based on SVM-RFE Feature Selection

针对支持向量机的可解释性,提出了一种基于SVM-RFE特征选择的规则提取方法。这一方法在预处理阶段采用优化的SVM-RFE来获取重要属性集,并设计和实现一种变型的顺序覆盖规则算法进行规则生成和裁剪,以兼顾可理解性与准确率和忠实度之间的平衡。仿真实验表明,这一方法准确率较高,产生的规则数量和条件项数也比较少。

With regard to the interpretation of result based on SVM,a novel rule extraction method based on feature selection SVM-RFE is presented.During pre-extraction phase,an improved SVM-based feature selection SVM-RFE is adopted to rank the importance of attributes,and a modified sequential covering approach is used to extract rules from the modified support vectors,which adopts some technical methods to coordinate the accuracy and fidelity with comprehensibility.Numerical experiments have shown the high accuracy and good comprehensibility of the proposed approach.