基于二进制PSO算法的特征选择及SVM参数同步优化

特征选择及分类器参数优化是提高分类器性能的两个重要方面,传统上这两个问题是分开解决的。近年来,随着进化优化计算技术在模式识别领域的广泛应用,编码上的灵活性使得特征选择及参数的同步优化成为一种可能和趋势。为了解决此问题,本文研究采用二进制PSO算法同步进行特征选择及SVM参数的同步优化,提出了一种PSO-SVM算法。实验表明,该方法可有效地找出合适的特征子集及SVM参数,并取得较好的分类效果;且与文4]所提出的GA-SVM算法相比具有特征精简幅度较大、运行效率较高等优点。

Simultaneous Feature Selection and SVM Parameters Optimization Algorithm Based on Binary PSO

Feature selection and classifier parameter optimization are two important aspects for improving classifier performance and are solved separately traditionally. Recently, with the wide applications of evolutionary computation in pattern recognition area, simultaneous feature selection and parameter optimization become possible and tendency. To solve the problem, we propose a simultaneous feature selection and SVM parameter optimization algorithm based on binary PSO algorithm called PSO-SVM. The experiments show that the algorithm can efficiently find the suitable feature subsets and SVM parameters, which result in good classification performance. Compared with GA-SVM , PSO-SVM can get a more compact feature subset and run more efficiently.