Recurrent Neural Network Based Cognitive Ability Analysis In Mental Arithmetic Task Using Electroencephalogram

Abstract

With the rapid increase in the application areas of Machine Learning and the requirement of cognitive ability detection, the use of Electroencephalogram has become a very effective tool to detect and record electrical activities in our brain. The visual inspection of the conventional methods of neurology can be a bit time consuming and affected by artifacts, which lead to inconsistent results later. The novelty in this idea behind cognitive ability analysis using mental arithmetic tasks lies here. In this paper, after collecting EEG data during mental arithmetic tasks performed by forty subjects, we have extracted features using a novel combination of power spectral density and correntropy spectral density method. The identification of cognitive ability using recurrent neural network has been carried out by classifying the subjects into two classes, i.e., good calculator and bad calculator. The bad calculators get asked to practice more for improving their performance in the next trial. The proposed approach outperforms the existing ones in the concerned domain in terms of its performance and it is well suited as it is flexible for all and the privacy of the user is also maintained.