An interpretable English reading proficiency detection model in an online learning environment: A study based on eye movement

Abstract

Aiming at the low English reading proficiency of ESL (English as second language) students in online learning environments, this study proposed an eye-movement-based machine learning monitoring model to detect English reading proficiency in real time. Eye-movement data from 43 students while completing online English reading tasks were recorded and 31 eye-movement features were extracted from the taxonomy of fixation, saccade, movement direction and gaze velocity. During the model training phase, LightGBM achieved an accuracy of 96.51 % in detection. An interpretable model, SHAP (SHapley Additive exPlanation), was used to explain the main effects of eye-movement features in detection, where high gaze velocity, absolute saccade direction, and average saccade duration were found to be strong indicators of English reading proficiency. Furthermore, SHAP analysis allows the identification of individual factors contributing to differences in English reading proficiency. This study demonstrated the effectiveness of the combination of eye-movement data with machine learning methods to identify students with low English reading proficiency in online reading and provided insights into the intrinsic correlation between eye movements and reading.