Random Forest Classification for Training a Brain Computer Interface (BCI)

1 Brain–computer interfaces (BCIs) aim at providing a non-muscular channel 2 for sending commands to the external world using brain activity. Most 3 existing BCIs detect specific mental activity in a so-called synchronous 4 paradigm. Unlike synchronous systems that are operational at specific 5 system-defined periods, self-paced interfaces have the advantage of being 6 operational at all times. Existing BCI systems rely on feature extraction 7 followed by a classification scheme to detect intentions from the brain 8 signal. In this paper, we propose a novel self-paced BCI system that 9 employs Random Forest (RF) algorithm for the classification of brain 10 signal. Unlike the conventional BCI systems, the proposed system does not 11 have a feature extraction step and tries to implicitly learn features from the 12 raw brain signals. We also employ a Bayesian optimization framework to 13 tune the parameters of the RF algorithm and the BCI system. The 14 performances of the proposed novel BCI system and a grid search method 15 are compared on dataset I of BCI competition IV. On the calibration data 16 our optimization method outperformed the grid search method by at least 17 11% accuracy. As expected, the results of both methods on the evaluation 18 dataset were not promising as the brain signal recordings in the calibration 19 and evaluation sessions followed two different paradigms 20