Applying Support Vector Machines with Different Kernel to Breast Cancer Diagnosis

Abstract

Breast cancer diagnosis is one of themajor problems in the medical field. It represents thesecond type of the largest cases of cancer deathsin women. Several techniques have been foundto solve the problem or make a better diagnosis.Recently, Support Vector Machine based systems arethe most common and are considered a better diagnosticassistant in cancer detection research. The qualityof the results generated depends on the choice ofsome parameters such as the kernel function and themodel parameters. In this paper, we analyze andevaluate the performance of several kernel functions inthe SVM algorithm. Experiments are conducted withdifferent training-test phases generated by the holdoutmethod and we used the WBCD (Wisconsin BreastCancer Database) to analyze the results. The resultsare evaluated by using the following performancesmeasures: classification accuracy rate, sensitivity,specificity, positive and negative predictive values. Tovalidate the results obtained by these different kernelfunctions, we use different values for the kernel functionsparameters and SVM model parameters and we recordthe optimal parameters values. Finally, we show thatthe Cauchy kernel and the Rational Quadratic kernel areidentical and converge to the same value.