This study investigates the predictive role of psychological constructs and contextual resources in shaping mathematics achievement and engagement using data from PISA 2012. Four outcomes were examined: mathematics performance, intentions, behavior, and subjective norms. Key predictors included mathematics self-efficacy, self-concept, anxiety, and household wealth as a proxy for digital access. Multiple linear regression and K-Nearest Neighbors (KNN) regression were applied to compare explanatory and predictive capacities. Results indicate that mathematics self-efficacy consistently enhanced achievement and engagement, while mathematics anxiety negatively influenced all outcomes. Household wealth significantly predicted test performance, but showed limited associations with motivational constructs. Across evaluation metrics, linear regression outperformed KNN, highlighting the continued relevance of interpretable statistical models in educational research. The findings contribute to ongoing discussions on digital equity and affective determinants of learning, underlining that both contextual resources and psychological dispositions are critical for mathematics success. Methodologically, the study illustrates the trade-off between the transparency of regression models and the flexibility of machine learning approaches, offering guidance for the integration of predictive analytics in education.