FEATURE SELECTION WITH RANKING SUPPORT VECTOR MACHINE VISUALIZATION

Abstract

In this paper, we first consider an application of the Nomogram visualization technique, which is a well-known one for describing numerical relationships in a graph, to ranking support vector machine. And then we utilize it to construct a feature selection method for ranking problems. In order to represent each feature on the log odds ratio in the nomogram, we use a probabilistic ranking support vector machine. Its purpose is to map the ranking support vector machine outputs into a probabilistic sigmoid function whose parameters are trained by using cross-validation. The effectiveness of our proposal helps the analysts study the effects of predictive features. Evaluation of the performance of ranking support vector machine visualization on the OHSUMED datasets shows that the proposed method is effective in feature selection.