Filter‑based feature selection in the context of evolutionary neural networks in supervised machine learning

Abstract

This paper presents a workbench to get simple neural classifcation models based on product evolutionary networks via a prior data preparation at attribute level by means of flter-based feature selection. Therefore, the computation to build the classifer is shorter, compared to a full model without data pre-processing, which is of utmost importance since the evolu tionary neural models are stochastic and diferent classifers with diferent seeds are required to get reliable results. Feature selection is one of the most common techniques for pre-processing the data within any kind of learning task. Six flters have been tested to assess the proposal. Fourteen (binary and multi-class) difcult classifcation data sets from the University of California repository at Irvine have been established as the test bed. An empirical study between the evolutionary neural network models obtained with and without feature selection has been included. The results have been contrasted with non parametric statistical tests and show that the current proposal improves the test accuracy of the previous models signifcantly. Moreover, the current proposal is much more efcient than the previous methodology; the time reduction percentage is above 40%, on average. Our approach has also been compared with several classifers both with and without feature selection in order to illustrate the performance of the diferent flters considered. Lastly, a statistical analysis for each feature selector has been performed providing a pairwise comparison between machine learning algorithms.