A new SVM-RFE approach towards ranking problem

Support vector machine recursive feature elimination (SVM-RFE) is a simple and efficient feature selection algorithm which has been used in many fields. Just like SVM itself, SVM-RFE was originally designed to solve binary feature selection problems. In this paper, we propose a new recursive feature elimination method based on SVM for ranking problem. As against standard approaches of treating ranking as a multiclass classification problem, our approach enables the use of standard binary SVM-RFE algorithms for ranking problems. We evaluate our algorithm on both public dataset and for a real world credit evaluating problem. The results obtained demonstrate the superiority of our algorithm over extended SVM-RFE to solve multiclass problems using ensemble techniques.