An Ensemble based Framework for Eye State Prediction from EEG data

Abstract

Since eyes are an important organ of human beings, eye gestures can be widely used in various smart systems, especially in identifying brain activity pattern. Electric signals are transmitted between brain cells for transfer of data, therefore using Electroencephalography (EEG), electrical activity of the brain can be recorded. With the help of small electrodes placed on human scalp, these electric signals can be captured. Based on this EEG, state of eye i.e. open or closed is predicted. Eye State Prediction has application in many fields such as driver drowsiness prediction using sequence of eye states, stress detection, military scenarios, etc. It plays a vital role in medical field also where it can be used as medium of communication for patients who are paralyzed or severely handicapped. It can also be used for controlling computer systems using eye gaze. In the previous work done in this field, data mining techniques and classifiers have been applied to the EEG dataset, however, hybridization (or ensembling) in addition to individual classifiers yields better results. In this study, we have developed an ensemble based model based on majority voting technique and compared its performance with different classifiers on the basis of various performance parameters such as accuracy, sensitivity, and specificity, and shown that ensemble based model performs better than other classifiers.