Brain Computer Interface for Eye Movement Recognition Using Random Forest

Expert system based on eye movement is of great benefit to wheelchair bound persons and amputees. Discrimination of the movements of eye is vital for EEG based controls. This paper proposes a system to classify four different eyeball movements i.e., left, right, up and down based on electroencephalography (EEG) signals. Eight subjects at the age of 21 performed the data acquisition task. The electrodes Fp1, Fp2, F3, F4, F7 and F8 are used for data acquisition. Sliding window technique is used along with pre-processing by notch filter (49–51 Hz) and band pass filter to eliminate artifacts and preserve the features. The signals in the frequency range 12Hz to 50 Hz are found to be contributing towards class identification. Random Forest, Decision Tree and Support Vector Machine classifiers are used in the proposed system to detect the horizontal and vertical eyeball movements. Among these classifiers Random Forest provided the highest accuracy of 91.1% for horizontal eye movements which consists of left and right movements and 87.1% for vertical eye movements which consists of up and down movements.

I. Introduction

The number of people suffering from neurological disorders have increased to one billion [1]. This people suffer from disorders around 6.8 million people worldwide suffer from neuroinfectious diseases, migraines, brain traumas, Parkinson's, Alzheimer's, and multiple sclerosis in addition to epilepsy, strokes, and multiple sclerosis. [1] per year falling prey to these conditions. The eyeball movements classified based on EEG data can be used to assist such people to increase their mobility. Brain-Computer Interface (BCI) use brain signals to control tasks like robotic limbs, wheelchairs and give paralysed people back some of their motor skills [2]. The majority of these high-tech devices for individuals with severe motor impairments are based on motor imagery or visual evoked potential. EEG signal generated while performing a particular movement contains sufficient data before a conscious decision to move any part of the body [3]. The obtained signals can be used to figure out how the features are related and to predict when a command will be carried out by the BCI system. In the proposed system, EEG signals from four different eyeball movements are analysed. EEG data is collected from 8 participants and the data is then used for preprocessing, feature extraction, and to classify the data, machine learning models are put into practice.