改进粒子群与支持向量机混合的特征变换

研究了数据挖掘中通过特征变换的数据预处理来提高支持向量机(SVM)分类精度的方法，提出了改进粒子群优化(PSO)和SVM混合的方法. 用推广t统计、Fisher判别式和随机森林的线性加权度量来排序特征，得到预选特征子集，再用启发式信息加速改进PSO搜索特征的线性变换因子，并用二进制PSO对特征变换子集进行特征选择，在后处理中通过格子搜索获取了高精度SVM分类器. 在NIPS 2003的madelon及10个UCI数据集上的实验表明，与有C-SVM分类精度相比，新方法在4个数据集上的精度更高.

Hybrid Feature Transformation Based on Modified Particle Swarm Optimization and Support Vector Machine

This paper researches on a preprocessing method of feature transformation in data mining to improve the classification accuracy of Support Vector Machine (SVM), and proposes an algorithm based on hybrid modified Particle Swarm Optimization (PSO) and SVM for transforming feature set in order to obtain higher accuracy classifier. First, the method preselects top-ranked features as feature subset using linear weighted combination of extended t-statistic, Fisher Discriminant Ratio (FDR) and Random Forests (RF) feature importance scores; secondly, uses modified PSO to search feature linear transformation factor and a novel heuristic info to accelerate it; then, selects feature subset on transformed dataset using Binary Particle Swarm Optimization (BPSO) algorithm to refine the transformed feature subset; last, uses Grid searching method to attain high accuracy SVM classifier. Experiments show the method proposed provides higher accuracy than the original SVM on the UC Irvine Machine Learning Repository (UCI) and NIPS2003 datasets.