International Journal of Applied Mathematics and Machine LearningAuthor's: Yan Deng
Pages: [13] - [22]
Received Date: November 10, 2017
Submitted by: Jianqiang Gao
DOI: http://dx.doi.org/10.18642/ijamml_7100121908
Based on practical examples, three kernel parameter optimization methods of Support Vector Machines (SVM), i.e., grid method, particle swarm optimization, and genetic algorithm, are compared. The results are as follows: (1) Grid method lasts the shortest time in optimization process, and as the number of samples increases, it shows more obvious advantages. (2) Optimization results of particle swarm optimization and genetic algorithm are both unstable, and take longer time on average. Due to their wider search scope, it is expected that more suitable kernel parameters can be obtained through many experiments. (3) With the increasing of the number of samples, kernel parameter optimization time will increase sharply. For large sample classification model, it is necessary to explore new SVM training algorithm for large-scale data set.
support vector machines (SVM), kernel parameter; parameter optimization.
