Prediction of students' stress and cognitive performance based on eye-tracking data

• Main Author: Lee, Cherng En
• Other Authors: Chen Chun-Hsien
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2022
• Subjects: Engineering::Aeronautical engineering
• Online Access: https://hdl.handle.net/10356/157882
• Institution: Nanyang Technological University

Stress levels amongst students have been rising in recent years due to the occurrence of Covid-19 global pandemic. Psychosocial stress is a common type of stress that can affect students’ mental health and cognitive performance. Impairment of cognitive abilities in terms of information processing and decision making can negatively impact students’ academic performance. Previous studies had utilised simple eye-tracking features to predict stress and assess cognitive abilities, but there has been lacking attempts on relating stress with cognitive performance. In this current study, we propose to predict psychosocial stress and cognitive performance of students using a wide range of eye-tracking features. First, a gradient boosting decision tree model, LightGBM (Light Gradient Boosting Model) was used as the main machine learning model to predict stress and cognitive performance. Second, to assess how each eye-tracking feature affects each prediction, SHapley Additive exPlanations (SHAP) values were used to identify the features with higher importance and its contributions in predicting stress and cognitive performance. Third and last, Analysis of Variance (ANOVA) test was conducted to investigate the interactive effect of stress together with cognitive performance. 44 participants were given 3 comprehension reading tasks each, with 1 in controlled conditions and 2 others with stress induction. Using eye-tracking data relating to fixations, saccades and gaze movements, the proposed model was the best performing when compared to other supervised machine learning models. Our study provided an insightful result on how stress and cognitive performance can be predicted by eye-tracking features, and how these two predicted subjects had interacted.