Classification of Motor Imagery Using Statistical Models

Nowadays, brain computer interfaces have become very popular in rehabilitation therapy, tools to help people with disabilities, daily application and videogames. There exist different models that use different types of brain signals. One of them is the model that employs sensorimotor rhythms, based on motor imagery. This work describes a semi-supervised method to classify motor imagery signals of the hand, foot, and a class that indicate inactivity, or relaxation. The classifier uses features obtained with statistical moments, and independent component analysis. Two classifiers were evaluated, one based on the k-means which achieved a performance of 85.16% during training, and 85.76
^%
in validation. The other classifier tested was based on the knn. This classifier had a training performance of 90.45% and 90.64% in validation.