Machine learning-based approach for identifying mental workload of pilots

In general, the fighter pilots are required to engage themselves entirely during flight operations involved in airto-air combat while in the cockpit of a fighter aircraft. The performance has to be monitored continuously by classifying their cognitive workload levels during different phases of flying. Towards this direction, an experimental study was conducted in a realistic high-fidelity flight simulator environment to classify the Pilots' Cognitive Workload (PCWL) level. A real-time implementation of algorithms to effectively organize the PCWL during takeoff, cruise and landing phases, physiological signals such as ECG and EEG of fighter pilots are used. The classification algorithms such as Linear Discriminant Analysis (LDA) classifier, Support Vector Machine (SVM) classifier, k-Nearest Neighbour (k-NN) classifier have been employed. It has resulted that takeoff (LDA - 75%, kNN - 60% and SVM - 75%) and landing phase (LDA - 75%, kNN - 60% and SVM - 75%) was better classified by HRV features while using PCA and cruise phase was classified better using EEG features (LDA - 72.44%, kNN - 62.92% and SVM - 59.02%) when PCA feature reduction technique was adopted. Using significant features by feature selection methods (PCA, statistically significant features) have shown improved classification accuracy compared to all the features classification method. The LDA and SVM are consistent classifiers compare to the kNN classifier. This study helps to classify the PCWL level at each flying phase due to increased task.