A Novel Framework for Cognitive Load Estimation from Electroencephalogram Signals Utilizing Sparse Representation of Brain Connectivity

Cognitive load is the quantity of mental activity imposed on a user’s working memory while performing any cognitive task. As the performance of a human depends on the imposed mental workload, estimating cognitive load is critical for maintaining human efficiency through cognitive monitoring. Recently, cognitive load estimation utilizing the brain signals recorded by Electroencephalogram (EEG) has gained popularity. Traditional EEG signal-based cognitive load detection methods generally focus on extracting temporal and frequency-based features from individual EEG electrodes, thus neglecting dynamic functional relationships between the brain regions. This study proposes a cognitive load estimation framework utilizing the sparse representation of the brain network data. From the multichannel EEG data, brain networks are obtained using functional connectivity measures, and the sparse codes of the same are computed using the Orthogonal Matching Pursuit algorithm. Then a sparse representation-based classifier is developed that utilizes the sparse codes. The proposed framework is implemented on a 3-class cognitive load dataset recorded at IIT Kharagpur, and the model’s efficacy is measured through sparse reconstruction error and classification performances.