Can Feature Selection be used to Detect Physiological Components in P300 based BCI for amyotrophic lateral Sclerosis patients?

The detection of brain state changes can dramatically improve the comprehension of cerebral functioning. To reach this aim, machine learning based automatic tools may be extremely useful to correctly classify different brain responses. The performance of these instruments depends on the features and the classification algorithm employed, but also from a good data preprocessing able to improve the poor signal-to-noise ratio [4] of the EEG signal. In this work, we combine data preprocessing with a feature selection based on the filter ReliefF and the linear SVM classifier LibLinear in order to analyse the data deriving from a P300 speller paradigm on patients with Amyotrophic lateral sclerosis (ALS). The purpose of this study is twofold: on the one hand we want to maximize the predictor’s performance, but most importantly, we aim at showing how the features ranking can be used to support scientific hypotheses or diagnoses.