Evaluation of Fatigue and Attention Levels in Multi-target Scenario using CNN

Fatigue is a behavioral phenomenon that occurs when a user conducts focused mental activity for a prolonged duration resulting in performance errors or lapse. This paper presents an electroencephalogram (EEG)-based fatigue and attention level detection using 2-dimensional convolutional neural networks (2D-CNN). The experimental paradigm was designed to identify and detect a target object from a multi- target scenario. The results show that our proposed model provides a classification accuracy of 86.12%, which is ~18%, ~16% and ~10% higher than the Bayesian linear discriminant analysis (BLDA), support vector machine (SVM) and bootstrap aggregating (bagging tree) algorithms. Further, the decreased session-wise accuracy levels observed for each subject after the 4th experimental session postulates cognitive state disparities were caused due to increased fatigue and dropped attention levels. These biomarkers were assessed by comparing the resting theta and alpha band powers with the rapid serial visual presentation (RSVP) performance of the later sessions (sessions 5 to 7). The results show an inverse relationship between the RSVP classification performance and the resting EEG power, validating that the subjects' performance is affected by the physiological state biomarkers like fatigue and attention for prolonged brain-computer interface (BCI) experiments. Our findings demonstrate that the resting theta and alpha band powers can be considered as indicative measures to interpret mental fatigue and attention deficit problems.