Bayesian approach for brain-computer interface using feature fusion techniques

Abstract

A 'brain-computer interface (BCI)' enables control of devices or communication with brain activity without using muscles. It has been successfully used in scientific, therapeutic applications and helps increase the patients' standard of life. 'Electroencephalography (EEG)' recorded from a person's scalp is used for controlling the 'BCI'. The major challenges of BCI are low signal-to-noise ratio of neural signals, and need of robustness of extracting feature set from the brain signals and classifying it. In this work, the authors review a data fusion techniques for 'EEG'-based 'BCI' along with Bayesian methods for 'BCI'. This paper provides a comparison of the feature extraction techniques - 'Laplacian (LAP)', Kalman and fused 'LAP'- Kalman. The features obtained were classified using Naive Bayes classifier. Source identification and spatial noise reduction is achieved through the surface 'LAP'. The two functions of surface 'LAP' are associated with prediction accuracy as well as signal orthogonality in 'BCI'.