Graph -Based Learning for Cognitive Load Prediction: A Novel Approach Using EEG and Graph Neural Networks

Cognitive Load Theory (CLT) emphasizes the limits of human cognitive processing, highlighting the need for effective learning strategies and workload assessment. Accurate cognitive load prediction enhances education, workplace performance, and human-computer interaction. Various machine learning models analyze cognitive load using EEG, which provides real-time brain activity monitoring. However, traditional machine learning models struggle to capture complex temporal patterns in EEG signals. This study proposes a novel approach using graph representation learning for cognitive load prediction using EEG signals. Unlike conventional methods, our approach constructs dynamic graph representations of EEG data, capturing temporal changes in cognitive states. We employ a Graph Isomorphism Network (GIN) to encode these graph structures into vector representations, which are then classified using Graph Neural Networks (GNN) and Deep Neural Networks (DNN). The GNN achieved a training F۱-score of ۰.۹۷۵۷, a cross-validation mean F۱-score of ۰.۹۴۷۳ ± ۰.۰۰۸۲, and a test F۱-score of ۰.۹۲۳۹, surpassing the DNN model, which recorded ۰.۹۴۱۷, ۰.۸۸۷۰ ± ۰.۰۱۷۸, and ۰.۹۰۵۳, respectively. The GNN also exhibited superior classification performance across cognitive load levels, achieving F۱-scores of ۰.۹۵۲۴ for mid-level, ۰.۹۳۳۳ for natural, and ۰.۹۱۱۸ for high-level cognitive load.

کلیدواژه ها:

Cognitive Load ، Arithmetic Tasks ، Stroop Tasks ، Graph Isomorphism Network ، Machine Learning