Twin Bounded Weighted Relaxed Support Vector Machines

Data distribution has an important role in classification. The problem of imbalanced data has occurred when the distribution of one class, which usually attends more interest, is negligible compared with other class. Furthermore, by the existence of outliers and noise, the classification of these data confronts more challenges. Despite these challenges, doing fast classification with good performance is desired. One of the successful classifier methods for dealing with imbalanced data and outliers is weighted relaxed support vector machines (WRSVMs). In this paper, the improved twin version of this classifier, which is called twin-bounded weighted relaxed support vector machines, is introduced to confront the mentioned challenges; besides, it performs in a significant fast manner and it is more accurate in most cases. This method benefits from the fast classification manner of twin-bounded support vector machines and outlier robustness capability of WRSVM in the imbalanced problems. The experimentally, the proposed method is compared with the WRSVM and other standard SVM-based methods on the public benchmark datasets. The results confirm the efficiency of the proposed method.