A hierarchical interval outranking approach to classify students’ dropout risk

Student dropout is a complex phenomenon driven by multiple, heterogeneous, and often interacting factors related to academic performance, socioeconomic conditions, engagement, and institutional context. While predictive machine learning models have been widely used to address this problem, they frequently lack transparency and decision-support capabilities required by educational policymakers. In this paper, we propose a hierarchical interval outranking approach to classify students according to their dropout risk, explicitly accounting for imprecision, interaction among criteria, and hierarchical structuring of decision factors. The proposed approach reformulates student dropout assessment as a multi-criteria ordinal classification problem, where students are assigned to ordered risk categories (e.g., high, medium, and low dropout risk). Criteria are organized in a hierarchy reflecting major dimensions of academic success, and criterion performances are modeled using interval data to capture uncertainty and variability inherent in educational records. Interaction effects such as synergy, redundancy, and antagonism between criteria are explicitly incorporated within the outranking framework. The classification is performed using hierarchical extensions of interval-based outranking methods, ensuring consistency, monotonicity, and interpretability of the results. The methodology is illustrated using a real-world dataset on student academic performance and dropout outcomes. Results show that the proposed approach provides interpretable and robust classifications, enabling partial analyses at different levels of the hierarchy and offering meaningful insights into the drivers of dropout risk. Compared to purely predictive approaches, the proposed method enhances transparency and supports informed decision-making for early intervention and academic policy design.