Hybrid Deep Learning Models for Accurate EEG-Based Cognitive State Classification

Hybrid Deep Learning Models for Accurate EEG-Based Cognitive State Classification

연구 분야: Infrastructure

학회: SN Computer Science

Accurately identifying cognitive states such as attention, interest, and mental effort is critical for applications in education, healthcare, and human-computer interaction. Traditional observational methods for assessing attention are often prone to human bias and subjectivity, underscoring the need for objective and automated approaches. Electroencephalography (EEG) offers a non-invasive, real-time method for capturing brain activity, making it a promising modality for this purpose. In this study, we propose a novel hybrid deep learning model that integrates multiple architectures to enhance the classification of cognitive states from EEG signals. The model was trained and validated using data collected from 18 participants engaged in attention-demanding tasks. It demonstrates the ability to reliably predict levels of attention, interest, and mental effort exerted by students during task performance. Both intra-subject and inter-subject evaluations were performed using five-fold cross-validation. Experimental results show that the Hybrid model attains an accuracy of 93% and 88% percent of intra and inter-subject classification, respectively, superior to any of the baseline methods in terms of accuracy, F1-score, and balanced accuracy. The analysis in terms of latencies also demonstrates that the specified architecture is feasible in terms of real-time under laboratory hardware limits. The outcomes show the potential and failings of deep learning in EEG-based cognitive state estimation, and indicate research directions in explainability, bias reduction, and cross-device generalizability. These findings highlight the potential of hybrid deep learning approaches for robust EEG-based cognitive state recognition, with implications for adaptive learning systems, neurofeedback applications, and cognitive workload assessment.