A0286
Title: Deep learning applications in mental workload classification
Authors: Serenay Cakar - Middle East Technical University (Turkey) [presenting]
Fulya Gokalp Yavuz - Middle East Technical University (Turkey)
Abstract: The $n-$back task paradigm provides rich temporal data, capturing nuanced insights into working memory and cognitive demand across different conditions. Introducing innovative perspectives, deep learning techniques are integrated to classify mental workload levels from n-back cognitive data with dense and sparse features. Our findings highlight the efficacy of the extreme Deep Factorization Machine (xDeepFM) model, validated through stratified 5-fold cross-validation. Compared to the baseline model for the 0- vs 1-back classification task, this approach yielded substantial enhancements: 67.50\% accuracy, 68.74\% sensitivity, 66.24\% specificity, and 68.48\% F1-Score. Notably, dense features comprising combinations of hemodynamic measures and experimental variables, with subjects as sparse features, contributed significantly to these improvements. Additionally, incorporating Principal Component Analysis (PCA) led to notable enhancements: 53.03\% accuracy, 98.03\% sensitivity, and 70.37\% F1-Score, particularly evident in classifying the 0- vs 1-back condition using the xDeepFM model. These outcomes underscore the significant role of deep learning methodologies in accurately classifying mental workload levels from complex $n-$back cognitive science data, providing invaluable insights into cognitive functioning and workload assessment.
