Classification of Pilots’ Mental Workload States Using Multi-Modal Physiological Signals and Machine Learning

Accurate assessment of pilots' mental workload is critical for aviation safety, as cognitive overload is a leading cause of accidents. Traditional methods, such as subjective scales (e.g., NASA Task Load Index) and behavioral observation, suffer from inherent biases and limited sensitivity to subtle cognitive changes. To address these challenges, we propose a data-driven framework for workload state classification using multi-modal physiological signals (EEG, ECG, EMG, EOG). The signals undergo preprocessing via filtering, normalization, and sliding window segmentation, followed by feature extraction (time-domain statistics, power spectral density, Riemannian kernels) and fusion. Three machine learning models—Support Vector Machine (SVM), Random Forest (RF), and Long Short-Term Memory (LSTM)—are implemented and optimized. Experimental results on a dataset of 109,000 samples demonstrate that the proposed approach achieves over 90% accuracy, outperforming traditional subjective methods. The RF model outperforms others, attributed to its ability to capture complex nonlinear relationships in the data and its strong fitting capability. This study validates the potential of multi-modal signal fusion and deep learning for real-time cognitive state monitoring in aviation.