A Framework for Cognitive Load Recognition Based on Machine Learning and Multimodal Physiological Signals by Wearable Sensors

Cognitive load recognition is of great significance to the study of emotional health, as excessive cognitive load not only impairs work and learning efficiency but also leads to mental pressure and fatigue, affecting overall well-being. Advances in wearable technology and understanding the role of physiological signals in assessing human health have enabled the possibility of cognitive load recognition using these signals. This work proposes a framework that combines multiple physiological signals (ECG, EDA, and PPG) collected through wearable sensors for cognitive load recognition. Heart rate variability (HRV) features are extracted from ECG signals, while statistical characteristics are extracted from EDA and PPG signals. The RFECV feature selection algorithm is utilized in the feature selection module to achieve optimal feature subset selection and redundant feature removal. The classification of cognitive load states is implemented based on the ASAPSO-LightGBM(Adaptive Simulated Annealing Particle Swarm Optimization-Light Gradient Boosting Machine) machine learning model. We validated the effectiveness of the framework on two publicly available datasets (MAUS and WESAD). Experimental comparisons of feature combinations and machine learning models demonstrate that the proposed framework achieves superior performance, with recognition accuracy of 98.52% and 88.38% on the two datasets, respectively.