Least squares support vector machine for class imbalance learning and their applications to fault detection of aircraft engine

Abstract Imbalanced problems often occur when the size of majority class is bigger than that of the minority one. The Least squares support vector machine (LSSVM) is an effective method for solving classification problem on balanced datasets. However, LSSVM has bad performance on minority class facing with class imbalance learning for the classification boundary skewing toward the majority class. In order to overcome the drawback, LSSVM for class imbalance learning (LSSVM-CIL) is proposed. LSSVM-CIL utilizes two different regularization parameters C + and C − that evaluate different misclassification costs. Furthermore, a method of combining reduced technique and recursive strategy is proposed to reduce the size of support vectors and retain representative samples. In addition, decomposition of the matrices via Cholesky factorization is employed as a solution to enhance the computational stability. Furthermore, the effectiveness of the two algorithms presented in this paper is confirmed with experimental results on various real-world imbalanced datasets. Fault detection of aircraft engine can be regarded as a CIL problem and has the demand for the real time. Finally, experiments on aircraft engine indicate that the two algorithms can be selected as candidate techniques for fault detection of aircraft engine.