Feature Extraction and Classification of Electroencephalogram Signals

Abstract

Brain Computer Interface (BCI) is a system that transforms human brain activities to control commands. It allows users to communicate with the external environment and reduce dependency on previous communication pathway of nerves and muscles. This system enable s patients suffering from partial or complete body paralysis by diseases such as amy otrophic lateral sclerosis , brainstem stroke or other neuromuscular diseases to relay assistive devices. This system needs signal - acquisition hardware that is safe, conveni ent, portable and able to perform in all environment. Brain Computer Interface is a computer based system that acquire brain signals , analyses and translates them into commands to operate output devices to carry out a desired action. Devices includes robotic arm control , electric wheel chair, games and others produced by generating commands through system. Electroencephalogra phy is a medical imaging technique that measures electrical activity generated by the brain . This technique is typically non - inv asive in nature with the electrodes placed along the scalp . Electroencephalogram signal s are contaminated by noise due to external environment and other reasons , thus may results in generation of wrong commands. Successful i mplementation of Brain Computer Interface system depends on efficiency of recording signal s , pre - processing, feature extraction , feature selection and classification of Electroencephalogram signal s . The purpose of pre - processing is to enhance signal to noise ratio of Electroencephalogram signals . The feature extraction method extracts the features for proper representation of mental tasks and motor imag ery tasks. The system then performs the feature selection process to select relevant features using ranking strategy. Selected f eatures ar e then classified using various machine learning techniques. In present work, Electroencephalogram signals studied for different mental and motor imaginery ta sks . As, EEG Signals are non - linear and non - stationary in nature therefore, time frequency represe ntation techniques used to extract information from signals. Features are extracted and classified for successful implementation of Brain Computer Interface system. Analysis using Fractal Dimension and Common Spatial Pattern is per formed to evaluate the b ehaviour of mental and motor imagery Electroencephalogram signal s . Classifiers are trained and tested to obtain maximum classification efficiency. Classification algorithm s such as Support Vector Machine, Artificial Neural Network and Random Forest are use d to discriminate different tasks.