Classification of Learners’ Attention Based on Facial Temperature

Abstract

Abstract This study investigates how facial temperature information from a Continuous Performance Test (CPT) experiment can be used to categorize learners’ attention. The study uses machine learning methods, including Logistic Regression (LR), KNearest Neighbors (KNN), and Support Vector Machine (SVM), to determine which model works best for this classification problem. The study aims to answer two main questions: first, which features—such as nose, forehead, cheek temperatures, and latency in temperature fluctuations—are most relevant for utilizing facial temperature to classify attention? Second, which machine learning model produces the best accuracy in this classification task? The study’s methodology comprises participant recruitment, thermal camera capture, CPT application creation, problem identification through literature review, and data gathering. Results show that adding temporal latency and cheek temperature greatly improves classification accuracy, and KNN is the most suitable fit model. The CPT was implemented, datasets were made publicly available, facial temperature extraction techniques were developed, and the study provided insights into the best feature sets and classifiers for attention classification. The study’s limitations include the size of the participant pool and the thermal camera’s capabilities, which point to potential areas for future research to improve attention classification techniques and broaden the study’s applicability to a wider range of demographics.