An Embedded Method for Feature Selection Using Kernel Parameter Descent Support Vector Machine
We introduce a novel embedded algorithm for feature selection,using Support Vector Machine(SVM)with kernel functions.Our method,called Kernel Parameter Descent SVM(KPD-SVM),is taking parameters of kernel functions as variables to optimize the target functions in SVM model training.KPD-SVM use sequential minimal optimization,which breaks the large quadratic optimization problem into some smaller possible optimization problem,avoids inner loop on time-consuming numerical computation.Additionally,KPD-SVM optimize the shape of RBF kernel to eliminate features which have low relevance for the class label.Through kernel selection and execution of improved algorithm in each case,we simultaneously find the optimal solution of selected features in the modeling process.We compare our method with algorithms like filter method(Fisher Criterion Score)or wrapper method(Recursive Feature Elimination SVM)to demonstrate its effectiveness and efficiency.
Feature selection Support vector machine Kernel function
Haiqing Zhu Ning Bi Jun Tan Dongjie Fan
School of Mathematics,Sun Yat-sen University,Guangzhou 510275,China Center for Urban Science and Progress,New York University,New York 10012,USA
国际会议
广州
英文
351-362
2018-11-23（万方平台首次上网日期，不代表论文的发表时间）