Reproduce “CLARE: Cognitive Load Assessment in REaltime with Multimodal Data” on OpenResearch

About this paper

Anubhav Bhatti, Prithila Angkan, Behnam Behinaein, Zunayed Mahmud, Dirk Rodenburg, Heather Braund, P.James Mclellan, Aaron Ruberto and 6 more

Researchers at Queen's University developed CLARE, a multimodal dataset for real-time cognitive load assessment featuring ECG, EDA, EEG, and gaze data from 24 participants, uniquely synchronized with self-reported cognitive load scores every 10 seconds. The study established benchmark binary classification results using various machine learning and deep learning models, showing a Transformer model achieved up to 85.58% accuracy in 10-fold cross-validation and 72.70% in leave-one-subject-out validation for distinguishing low from high cognitive load.

Key insights

• The Multi-Attribute Task Battery (MATB-II) effectively induced diverse subjective cognitive load experiences across participants, as evidenced by the wide distribution of self-reported scores.
• An 'inverted S' pattern was observed between increasing task complexity and reported cognitive load, with most participants experiencing moderate load at average or higher complexities.
• Deep learning models, particularly the Transformer, generally achieved superior performance over classical machine learning methods in cognitive load classification, especially under the more challenging Leave-One-Subject-Out cross-validation scheme.

Summary by alphaXiv.