South African universities continue to struggle with low graduation rates and high dropout rates, primarily due to socioeconomic inequalities and inadequate school preparation. Traditional cognitive metrics, such as the Admission Points Score (APS), often fail to capture the full range of a student’s potential, particularly in the context of educational disparity. This chapter explores the predictive value of non-intellective factors (such as learning strategies and student wellness) by integrating data from two diagnostic instruments, the Learning and Study Strategies Inventory (LASSI) and the Wellness Questionnaire for Higher Education (WQHE), into a machine learning framework for early risk detection. Drawing on data from 6300 first-year students across the Humanities, Education, Management Sciences, and Economics and Finance faculties at a large South African university, the study employed an ensemble of classifiers configured as a voting system to predict academic risk, defined as delayed graduation. The results revealed that non-intellective factors, particularly self-regulated learning behaviours and wellness indicators, are potent predictors of educational underperformance. Models that combined APS with LASSI and WQHE performed the best. Notably, even models excluding APS achieved comparable accuracy, underscoring the independent predictive strength of non-intellective data. The findings advocate for a paradigm shift toward holistic, data-informed student support systems. Integrating instruments like LASSI and WQHE into early warning systems can surface at-risk students who might otherwise be overlooked. Furthermore, targeted interventions tailored to individual learning and wellness profiles, ranging from academic coaching to mental health support, can enhance retention and equity. The chapter concludes with a call to expand learning analytics initiatives within a responsible AI framework to support student success in complex environments.