C-Support Vector Classification: Selection of kernel and parameters in medical diagnosis

This paper investigates the impact of kernel function and parameters of C-Support Vector Classification (C-SVC) to solve biomedical problems in a variety of clinical domains. Experimental results demonstrate the effectiveness of optimizing parameters for C-SVC with different basic kernel. Without optimizing parameters results for classification accuracy with data sets in medical domains shows the best performance of linear kernel. After optimization of parameters, results of classification accuracy are more consistent for all kernel functions, and we no longer have the dominance of certain kernel functions, or larger variance in the results. The biggest benefits of optimization had those kernel functions, which have a smaller accuracy of classification. Results show that time taken to build model are very high with C-SVC and polynomial kernel, compare with others kernels.