Workload Analysis of Pilot Steep Turn Maneuvers Using SR20 Aircraft and EEG Data

This study aims to analyze the workload differences between left and right turns in steep turn maneuvers among pilots using electroencephalogram (EEG) data. Thirty-seven flight cadets were recruited for the experiment, utilizing an SR20 aircraft desktop flight simulator. EEG data during steep turn tasks were recorded using the Emotiv EPOC Flex 32 system, and a total of 800 features were extracted from 32 electrodes, including time-domain, frequency-domain, and nonlinear features. Significant differences in EEG features were observed between left and right turn tasks, particularly in the frontal and temporal regions. During left turns, the left hemisphere exhibited higher high-frequency energy and complexity, while during right turns, the right hemisphere showed predominant energy in the theta/alpha frequency bands. Six machine learning classifiers (XGBoost, LightGBM, GB, SVM, LR, and Linear SVC) were employed with all EEG features to identify workload differences between left and right turns. A total of 800 different EEG features were selected and ranked by importance using a random forest algorithm. Features were selected at proportions of 20%, 40%, 60%, 80%, and 100% as inputs for the six classifiers. The GB model achieved optimal performance at 20% feature proportion, with an accuracy of 91.02%, a receiver operating characteristic area under the curve of 0.9653. These findings contribute to demonstrating the effectiveness of EEG-based feature analysis combined with machine learning in distinguishing workload differences between left and right steep turns.