Workload analysis of pilot steep turn maneuvers using SR20 aircraft and EEG data

To compare left vs. right steep turns in terms of workload-related neurophysiological signatures using electroencephalogram (EEG) and machine learning.

Thirty-seven flight cadets performed one left and one right steep turn in an SR20 desktop flight simulator while a 32-channel EEG (Emotiv EPOC Flex 32) was recorded. From 2-s sliding windows (50% overlap), 800 features per window were extracted (time-, frequency-, and non-linear domains). Six classifiers (XGBoost, LightGBM, GB, SVM, LR, and Linear SVC) were evaluated using cross-subject nested cross-validation with variance-ranked feature subsets (20%, 40%, 60%, 80%, and 100%), and an additional 10% subset was assessed to identify a more parsimonious feature set.

LightGBM demonstrated superior performance across all feature proportions.

NASA-TLX was significantly higher in right turns than in left turns (5.55 ± 1.13 vs. 4.98 ± 1.06, p < 0.001, and Cohen's d = 0.52). Post-hoc interpretation combined RF-based importance and variance-ranked top-feature analysis, showing convergent frontal/frontocentral dominance with complementary utility definitions (predictive contribution vs. signal dispersion). Physiologically, left turns were associated with relatively higher high-frequency activity/complexity, whereas right turns showed relatively stronger theta/alpha-related patterns.

These findings support MWL-associated directional neurophysiological differences in steep turns and identify candidate EEG markers for lightweight real-time workload monitoring, facilitating optimized flight training and enhanced aviation safety.