Feature selection algorithms to reduce processing time in classification with SVMs

Abstract

By applying feature selection algorithms, such as the Relief and the Sparse Multinomial Logistic Regression with Bayesian regularization (SBMLR) to a feature set, a smaller subset of features can be obtained. Considering only those selected for all or most of the test subjects; this shows that the Mean Absolute Value (MAV) of the signal provides less information than the rest of the features that were selected. The proposed method was applied to the classification of myoelectric signals of the transtibial section, using Support Vector Machines (SVM) as a classifier.