Predicting ESL Learners’ Enjoyment Using Deep Learning at a Pakistani University

Abstract

Higher education across the globe has seen a significant transformation in language learning experiences due to the fast implementation of artificial intelligence (AI) tools into English as a Second Language (ESL) instruction. However, while there is mounting evidence that DL-supported learning improves motivation, engagement, and emotional experiences, little research has examined whether patterns of deep learning engagement can predict learners’ enjoyment. In order to fill this gap, the current study used a deep learning methodology to predict the enjoyment of ESL learners among undergraduate students at The University of Lahore (UOL), Pakistan, based on AI engagement characteristics. Data were gathered using a cross-sectional study design using self-report measures of deep learning engagement and enjoyment, such as behavioral usage, interaction depth, feedback engagement, and perception-based markers like trust in AI and autonomy support. Nested cross-validation was used to train and assess several supervised machine-learning models, including Elastic Net regression, Random Forest, and Gradient Boosting. Model interpretability was assessed using SHAP values to pinpoint important predictors. With non-linear models beating linear baselines, the results showed that AI engagement variables well predicted learners’ enjoyment. Notably, the greatest predictors were perceived autonomy support, feedback uptake, voluntary AI use, and interaction depth. Overall, by emphasizing the potential of predictive analytics for comprehending emotional outcomes and providing pedagogical insights for ethically responsive and emotionally supportive AI-integrated instruction in Pakistani higher education, the study adds to the growing body of literature on AI-mediated ESL learning.