The Effects of Kernel Functions and Optimal Hyperparameter Selection on Support Vector Machines

Authors : Aslı YAMAN, Mehmet Ali CENGİZ

Pages : 64-71

View : 16 | Download : 8

Publication Date : 2021-03-30

Article Type : Research Paper

Abstract :Support Vector Machine insert ignore into journalissuearticles values(SVM); is a supervised machine learning method used for classification and regression. It is based on the Vapnik-Chervonenkis insert ignore into journalissuearticles values(VC); theory and Structural Risk Minimization insert ignore into journalissuearticles values(SRM); principle. Thanks to its strong theoretical background, SVM exhibits a high performance compared to many other machine learning methods. The selection of hyperparameters and the kernel functions is an important task in the presence of SVM problems. In this study, the effect of tuning hyperparameters and sample size for the kernel functions on SVM classification accuracy was investigated. For this, UCI datasets of different sizes and with different correlations were simulated. Grid search and 10-fold Cross-Validation methods were used to tune the hyperparameters. Then, SVM classification process was performed using three kernel functions, and classification accuracy values were examined.
Keywords : Support vector machines, kernel function, tune parameter