Evidential SVM for binary classification

Support Vector Machine (SVM) is one of the most popular supervised learning methods for its better performances over diversified applications. SVM constructs a maximum-margin hyperplane and predicts the class of a new incoming data point based on that hyperplane. However, the hyperplane may not give a reliable decision in some cases specially when the training data is imprecise and noisy. To improve this situation, we propose an evidence based binary SVM classifier (EbSVM) where we first identify the sources of information (SoI) and propose a method to generate mass value from these SoIs. Finally, we combine these mass values using DS theory of evidence. An experiment over six benchmark datasets illustrates that EbSVM significantly performs better than the existing state-of-the-art methods.