Personalizing kernel and investigating parameters for the classification with SVM
Gend Lal Prajapati, Arti Patle · 2017
Investigation of kernel parameters and formulation of kernel in Support Vector Machine are recent research objectives. Parameter and kernel selections are most of the time problem dependent; the most common method for parameter selection is grid search. This paper investigates the parameters and formulates the personalized kernel Polyrobiaf for the SVM classification accuracy. SVM kernel and parameter evaluations are very helpful for many domains in classification. Evaluation represents the impact of parameter values on SVM classification accuracy with different kernels. Methodology concludes with the benchmark datasets from UCI repository. Results identify that the kernel parameter values are more effective for Classification accuracy and analyze how the kernel parameters affect classification performance. After evaluating the parameters, personalized kernel has been contrived. New kernel Polyrobiaf is combination of polynomial and RBF with respect to their qualities. Research is finalized by the experimental results after comparing the classification accuracy on the medical domain datasets. Validation is represented by graphically comparing the polynomial, RBF and new personalized kernel.