Feature Selection Techniques in Uterine Electrohysterography Signal

Abstract

To classify between labor and pregnancy contractions, feature extraction from uterine electrohysterography (EHG) signal has been used by many researchers. The number of features (linear and nonlinear) becomes huge when using them for classification techniques. The aim of this paper is to reduce the number of features and use only the pertinent ones to achieve the classification problem. In this paper, we compare the results of three selection methods. The first one is based on measuring the distance between the feature histograms of the 2 studied classes (pregnancy and labor). The second one is the Sequential Forward Selection method (SFS) and the third method is based on the Binary Particle Swarm Optimization techniques (BPSO). All methods give common pertinent features whatever the type of classifiers. These features will be used in future works to classify between labor contractions and pregnancy contractions.