Key Predictors of Academic Performance (PAP) Using Motivation, Social Support, and Institutional Quality: A Machine Learning Approach

Comprehending the complex factors influencing academic success is essential for formulating effective educational policy. Although current research has analysed individual elements like motivation and teaching quality, there is a scarcity of studies that incorporate these characteristics with advanced predictive models in Nigerian higher education. This study fills the gap by integrating conventional statistical techniques with machine learning ensembles to simulate the determinants of Grade Point Average (GPA). Data were gathered from 308 undergraduates at four Nigerian universities using a standard questionnaire. The investigation employed Ordinary Least Squares (OLS) regression for clarity and ensemble algorithms, namely Random Forest and Extreme Gradient Boosting (XGBoost), for predicting accuracy. The findings demonstrate that motivation is the most significant predictor of GPA, succeeded by family support and instructor quality. The XGBoost model exhibited superior predictive accuracy, achieving an R² of 0.9996 and a Mean Absolute Error (MAE) of 0.0047, thereby capturing intricate non-linear interactions more proficiently than linear models. The remarkably high accuracy indicates the necessity for careful interpretation of potential overfitting and necessitates further validation on larger, external datasets. These findings underscore the imperative of comprehensive educational techniques that cultivate intrinsic drive and enhance social support systems.