<p>This study investigates algorithmic fairness in higher education, focusing on predictive models designed to identify students at risk of academic underperformance. It explores the ethical and practical implications of biased predictions and examines whether imperfect models can still support equity when coupled with effective interventions. Using a student success model as a case study, the analysis covers a range of fairness metrics and mitigation strategies, including pre-processing, in-processing, and post-processing techniques. Results indicate that while algorithmic bias can be reduced, the effectiveness of mitigation approaches is highly context-dependent and shaped by specific fairness objectives. A real-world impact analysis reveals that even partial fairness improvements in model predictions can lead to meaningful gains for disadvantaged groups. These findings highlight the importance of balancing fairness metrics with tangible student outcomes. The study offers practical guidance for data scientists, educators, and policymakers, advocating for transparent, context-aware, and ethically grounded algorithmic practices to advance equitable student success.</p>