Abstract:
Brain‘s electrical activity recorded as Electroencephalography Signals (EEG) opens window to explore brain‘s functioning. EEG is a current focus of the modern research to detect the human‘s perception state, classify brain states, develop predictive systems, progress in medical diagnosis and military operations. Owing to the wide ranged benefits, efforts need to be put in to develop better feature extraction, selection and classification techniques for EEG.
A fourfold meticulous and efficient process has been introduced in this research. Data is pre-processed to eradicate fallacious terms. A novel concoction of new features based on statistical/ mathematical models and frequency analysis methods has been established. Data aspects that majorly contribute to distinguish the classes are sorted through application of advance feature selection techniques LASSO and Chi-Square scoring. Classification of brain waves has been performed through the seven well reputed machine learning algorithms. Two flavors of neural nets i.e. conventional neural network and the Deep Neural Networking, classification via hyper plane based Support vector machines, instance based K-Nearest Neighbor, two distinguishing decision tree rooted predictive models and logistic regression have been implemented. Towards the end research evaluates performance and compares the achievements in contrast to the previously published research works. It has been successfully demonstrated that the introduced methodology has improved the brain signal based eye state classification accuracy to 99% with the enhancement of 45% in as compared to raw data and has effectively reduced the computation time from two hours to few seconds. The research recommends KNN and Deep learning as a better choice for EEG data.
