A New Fuzzy Set and Nonkernel SVM Approach for Mislabeled Binary Classification With Applications

This paper proposes a new approach based on the kernel-free quadratic surface support vector machine model to handle a binary classification problem with mislabeled information. Unlike the traditional fuzzy and robust support vector machine models that reduce the weights of suspectable mislabeled points or even discard them, our new method first adopts the intuitionistic fuzzy set method to detect those suspectable mislabeled points, then deletes their labels, and indiscriminately utilizes their full position information to build a semisupervised model. In this way, we can not only eliminate the negative effect of mislabeled information but also avoid the difficult task of searching proper kernel functions in classical SVM models. Besides, to improve the efficiency and accuracy, a branch-and-bound algorithm is designed to accelerate the solving process. After that, we conduct some numerical tests with both artificial and real-world datasets to verify the superior performance of our proposed method among several benchmark methods. Furthermore, the proposed method is applied to brain-computer interface and credit risk assessment. The promising results strongly demonstrate the effectiveness of our method and show its big potential in some real applications.