Training Robust Support Vector Machine Based on a New Loss Function

摘要

To reduce the influences of outliers on support vector machine (SVM) classification problem, weconstructed a new tangent loss function. Since the tangent loss function is not smooth in some interval, asmoothing function was used to approximate it in this interval. According to this loss function, we got thecorresponding tangent SVM. The experimental results show that tangent SVM is less sensitive to outliersthan conventional SVM. So the proposed new loss function and tangent SVM are both effective.