Key points are not available for this paper at this time.
This paper presents a systematic review of machine learning (ML) methods for cognitive load (CL) assessment and classification. From 1,949 studies retrieved from the Scopus database, 43 papers were identified that employed ML approaches for CL classification. Unimodal moddels using physiological (i.e., EEG, fNIRS, and HRV) or eye-tracking (i.e., pupil dilation) features achieved classification accuracies between 40 to 96%, while multimodal approaches integrating eye-tracking, and physiological features reported up to 98% accuracies. This review categories various ML algorithms, ranging from traditional methods such as Support Vector Machines (SVM) to advanced neural networks. Ensemble methods (e.g., Random Forest and Gradient Boosting), and neural architectures including CNNs and LSTMs, consistently demonstrated superior performance, reaching accuracies of up to 99%, reflecting their capacity to model nonlinear patterns and capture temporal dynamics in physiological and eye-tracking data. This study compares methodological trade-offs and demonstrates the value of multimodal data fusion for enhancing CL classification accuracy.
Molloy et al. (Mon,) studied this question.