Using machine learning algorithms to predict students' general self-efficacy in PISA 2018

Self-efficacy is a critical psychological construct that exerts a positive impact on students' learning experiences and global well-being. Previous studies explored the factors related to the development and variation of students' self-efficacy, but they only focused on a limited set of predictors. To gain a more comprehensive understanding of the factors affecting self-efficacy, it is necessary to build a predictive model based on a large number of predictors using a data-driven approach. Therefore, guided by socio-ecological theory, we categorized 256 candidate predictors from the PISA 2018 student and school questionnaires in five levels of socio-ecological systems. We then used two machine learning algorithms, Lasso and XGBoost, to predict self-efficacy of 612,004 students aged 15 to 16 years from 79 countries and regions. The results showed that XGBoost outperformed Lasso. We then extracted feature importance from the best-performing XGBoost model to rank the features both overall and within each level of the socio-ecological systems. The analysis revealed that individual-level attributes such as mastery goal orientation, meaning of life, and positive emotions were the most important predictors of self-efficacy. Other significant contextual factors included parents' emotional support, home possessions, and school climate factors (e.g., cooperation climate). Furthermore, self-efficacy varied significantly across countries. This study advances our understanding of self-efficacy by identifying the important predictors from different levels of socio-ecological perspectives. The results suggest that self-efficacy is a composite outcome shaped by a myriad of influences spanning from individual factors to broader socio-ecological perspectives.