Human-Machine Interaction: EEG Electrode and Feature Selection Exploiting Evolutionary Algorithms in Motor Imagery Tasks

Brain Computer Interfaces (BCIs) based on the recording of electroencephalographic signals have revolutionized the human-machine interaction. Being in presence of heterogeneous electrophysiological data, that come with a low number of instances and a great number of features, it is necessary to find a solution that can achieve good performances with respect to all the subjects, having as input a restricted feature subset. Firstly, we propose a population-based approach that allows to mitigate the data heterogeneity. Secondly, not wanting to make assumptions on the feature types, we propose the application of genetic algorithm, particle swarm optimization and simulated annealing as evolutionary feature selection techniques. We present the results of our proposal on a motor movement/imagery experiment. From these results, we verified that each feature type contributes differently depending on the task and on the sensor it was computed on, thus giving a broader view of the different type of analyses that can be performed to allow a better interaction between a user-centric system like a BCI based on motor imagery and its human user.