Applying Support Vector Machines with Different Kernel to Breast Cancer Diagnosis

The detection of breast cancer poses a significant challenge in the field of medicine. It represents the second type of the largest cases of cancer deaths in women. Several techniques have been found to solve the problem or make a better diagnosis. Recently, Support Vector Machine based systems are the most common and are considered a better diagnostic assistant in cancer detection research. The quality of the results generated depends on the choice of some parameters such as the kernel function and the model parameters. In this paper, we analyze and evaluate the performance of several kernel functions in the SVM algorithm. Experiments are conducted with different training -test phases generated by the holdout method and we used the WBCD (Wisconsin Breast Cancer Database) to analyze the results. The results are evaluated by using the following performances measures: classification accuracy rate, sensitivity, specificity, positive and negative predictive values. To validate the results obtained by these different kernel functions, we use different values for the kernel functions parameters and SVM model parameters and we record the optimal parameters values. Finally, we show that the Cauchy kernel and the Rational Quadratic kernel are identical and converge to the same value.