An Ensemble-Based Feature Selection Algorithm Using Combination of Support Vector Machine and Filter Methods for Solving Classification Problems

A new feature selection algorithm for solving classification problems is proposed. The algorithm exploits the ensemble-based methodology and iteratively combines classifiers in order to assign weights to features characterizing their importance in classification. The algorithm is based on the joint use of a filter method and the well known support vector machine. Moreover, the filter method uses only support vectors instead of the total training set to calculate the feature weights. Numerical experiments with publicly available data sets show that the proposed algorithm improves the classification accuracy.