Evaluation of imbalanced datasets using fuzzy support vector machine-class imbalance learning (FSVM-CIL)

Abstract

Support Vector Machines (SVMs) work effectively for balanced datasets. While considering imbalanced datasets, it produces suboptimal classification models which is sensitive to outliers and noise present in the datasets. Fuzzy SVMs (FSVMs), which is a variant of the SVM algorithm also suffers from the problem of class imbalance. Hence in this paper we present a method to improve FSVMs for Class Imbalance Learning (CIL) i.e. FSVM-CIL, which handles the problem of outliers and noise. In this method we assign fuzzy membership values for training examples to reduce the effect of outliers and noise under the principle of cost-sensitive learning. In our proposed method we compare the results obtained by different existing imbalance-learning methods for the different datasets. Simulation results show that the proposed FSVM-CIL method produces a consistent solution for imbalance datasets by reducing the problem of outliers and noise. FSVM-CIL overcomes the disadvantages of SVMs in many practical applications especially in medical field like classification of multiple cancer types.