Channel selection for monovariate analysis on EHG

A multichannel system was used in order to record simultaneous EHG at different locations on the woman's abdomen. Using all channels leads to a large dimension of data for EHG processing. The aim of this paper is first the selection of the best channel(s) that provide the most useful information to discriminate between pregnancy and labor classes. For this purpose, the Relieff method is used in order to select the best discriminant channels. Then, after channel selection, a feature selection method named binary particle swarm optimization is used to select the best features (from the linear and nonlinear features used in this study) from the selected channels. This step is very important to facilitate classification problem. Additionally, the aim of this paper is to compare the results obtained by using monopolar and then bipolar EHG signals.