Mental Workload Estimation from EEG Signals Using Machine Learning Algorithms.

Multitasking conditions prevalent in many environments such as critical operations in defense activities, evaluating user interfaces in man-machine interaction, etc. require assessment of mental workload of operators. However, mental workload (MWL) cannot be perceived directly as it is a complex and abstract property of human physiology. The techniques available in the literature for its assessment usually depend on subjective analysis, performance analysis and psycho-physiological measurements. But, these approaches despite being followed often prove to be inadequate due to high inter-personal variations and inconvenient procedures and subjective to the bias of evaluators. With the recent advancements of proliferation of Brain Computer Interface (BCI) devices and machine learning algorithms, it is possible to estimate MWL automatically. Nevertheless, there is a need to address issue like managing a high dimension and high volume data in real time. In this work, we propose an approach to estimate the mental workload using electroencephalogram (EEG) signals of an operator while in operation. A thorough investigation of different features, optimization of features and selecting an optimal number of channels are the some of the crucial steps have been addressed in this work. We propose a novel feature engineering method to extract a reduced set of features and utilize only a sub-set of channels for the purpose of classification of workload into different levels with the help of supervised machine learning techniques. Further, we investigate the performance of different classifiers and compare their results. It can be inferred from the observed results that mental workload estimation using machine learning algorithms is a better solution compared to the existing approaches.