Abstract
The need for a practical framework capable of detecting the cognitive load level of users in fast-changing environments such as adaptive learning platforms, AI-enabled communication tools, and virtual reality environments systems is on the increase. Hence, this study offers a multimodal approach to cognitive load levels detection using brainwave signals (elec-troencephalography (EEG)) and motion data. The data comprises of five frequency bands including Delta, Theta, Alpha, Beta, and Gamma and 3-axis accelerometer and gyroscope measurements, with 52 features in total including EEG only (20), motion only (6) and the fusion of both (26). Cognitive load level labels were obtained by categorising the load levels into Low, Medium, and High tiers using quantile thresholds. Temporal features were obtained using 50-sample windows with 25-sample overlap, allowing for the computation of features for real-time use. Three techniques were used including Random Forest, Support Vector Machine (SVM), and Extreme Gradient Boosting (XGBoost), from the result multimodal features constantly performed better than single modal inputs (EEG only or Motion only) with XGBoost achieving the highest accuracy of 93.43%, Random Forest (91.2%), and Support Vector Machine (88.7%). The Receivable operating characteristics curve (ROC-AUC) for each of the models were more than 0.96%, showing a strong discriminative power of the multimodal feature set. The result has validated that EEG signals when combined with motion sensor data can provide a robust solution for real-time cognitive load level monitoring in adaptive systems, in education, and even in rehabilitation training. This study adds to digital healthcare, where cognitive workloads can be monitored using wearable physiological sensor data, and this can be applied mental fatigue monitoring, adaptive healthcare systems, and clinical workload assessment.