Abstract
Academic performance in higher education is influenced by both psychological conditions and teaching environments. To examine their combined effects, this study develops a data-driven, machine learning-based framework using five dimensions: learning motivation, academic self-efficacy, emotional instability, classroom engagement, and perceived teaching support. Based on data from 612 undergraduates, the study applies descriptive statistics, correlation analysis, multiple linear regression, Random Forest, and XGBoost to explore variable relationships and compare predictive performance. The results show that academic self-efficacy, classroom engagement, and perceived teaching support positively influence academic performance, whereas emotional instability has a negative effect. Classroom engagement and self-efficacy are the most influential factors. Machine learning models outperform linear regression in prediction, indicating the presence of nonlinear effects. The findings provide useful evidence for teaching improvement and student support.