Radial Basis Function Integrated with Support Vector Machine Model for Breast Cancer Detection

There is an urgent demand for the development of accurate prediction methods due to breast cancer being one of the most common diseases among women. Different statistical and machine learning methods have been used to develop predictive models for breast cancer. In particular, SVMs have been shown to outperform many related techniques. There are different kernel functions used while building an SVM classifier, and there is the likelihood that one will be better than the others in terms of prediction performance. But few studies have investigated a systematic assessment of the predictive accuracy based on various kernel functions used by SVM. In addition, the performance of SVM classifier ensembles that has been suggested as a way to improve classification results is yet unknown in breast cancer prediction. This study aims to assess comprehensively the predictive ability of SVM and its ensembles in both small/medium sized and large-scale breast cancer datasets. The evaluation involves several metrics, which include accuracy performance metric for classification task; receiver operating characteristic (ROC) analysis; F-measure and computational times spent on training the SVM and its ensembles. The experimental research reveals that the SVM ensembles, particularly those linear kernels with bagging implementation and RBF kernels using boosting method are preferred for miniature datasets. In such situations, feature selection is spotlighted during data preparation. On large datasets, RBF kernel-based SVM ensembles with boosting outperform other classifiers. This study provides important insights into improving SVM and ensemble methods for better breast cancer prediction, irrespective of the scale.