Predictive Modeling of Student Academic Risk in Conflict-Affected Zones: A Case Study Using Decision Trees in Higher Education Institutions of North Kivu (DRC)

Abstract

Higher education institutions in North Kivu, Democratic Republic of Congo (DRC), continue to operate under conditions of protracted conflict, marked by student displacement, infrastructural degradation, and psychological trauma. These challenges hinder academic performance and increase dropout rates, yet institutions often lack systematic tools to identify and support students at risk. This study examines the potential of predictive modelling using decision tree algorithms to identify students at academic risk in such fragile contexts. Using data collected from 350 students in conflict-affected institutions, key academic and socio-demographic variables were analyzed through RapidMiner. The resulting model achieved a classification accuracy of 99.05%, with substantial precision and recall scores across both classes. Psychological support, GPA, motivation level, and displacement status emerged as significant predictors of the outcome. The findings suggest that even in resource-constrained and unstable environments, accessible AI tools can support timely interventions and institutional decision-making. This research contributes to the discourse on educational resilience in conflict zones. It aligns with Sustainable Development Goals 4 and 16, highlighting the role of AI in fostering inclusive, equitable, and quality education in post-crisis settings.