Mental Workload Classification Using Bidirectional LSTM Networks with Multi-Feature Fusion

Mental workload assessment is crucial for optimizing human performance in various domains including brain-computer interfaces (BCI) and cognitive load monitoring systems. This paper presents a novel approach for mental workload classification using Bidirectional Long Short-Term Memory (BiLSTM) networks with comprehensive feature extraction from EEG signals. Our method incorporates multiple feature types including spectral power analysis, entropy measures, and Hjorth parameters to create rich temporal sequences for classification. We evaluate two BiLSTM architectures - basic and advanced - on the STEW dataset containing high and low mental workload conditions from 48 subjects. The advanced BiLSTM model achieves exceptional performance with$\mathbf{9 8. 5 0 \%}$5-fold accuracy and 98.40 % test accuracy, demonstrating the effectiveness of temporal sequence modeling for mental workload classification. The proposed approach shows significant improvements over traditional machine learning methods and provides a robust classification framework suitable for practical applications.