Generalized Fisher score for feature selection 论文

摘要

Fisher score is one of the most widely used su-pervised feature selection methods. However, it selects each feature independently accord-ing to their scores under the Fisher criterion, which leads to a suboptimal subset of fea-tures. In this paper, we present a generalized Fisher score to jointly select features. It aims at finding an subset of features, which max-imize the lower bound of traditional Fisher score. The resulting feature selection prob-lem is a mixed integer programming, which can be reformulated as a quadratically con-strained linear programming (QCLP). It is solved by cutting plane algorithm, in each it-eration of which a multiple kernel learning problem is solved alternatively by multivari-ate ridge regression and projected gradient descent. Experiments on benchmark data sets indicate that the proposed method out-performs Fisher score as well as many other state-of-the-art feature selection methods. 1