Generative AI-enhanced interventions: a novel framework for predicting and mitigating freshman student attrition

Student attrition remains a significant challenge for higher education institutions, particularly during the freshman to sophomore year transition. This study introduces a comprehensive decision-support framework that integrates state-of-the-art predictive machine learning (ML) techniques, local ML explanation techniques, and Generative AI to enhance individualized intervention programs aimed at reducing freshman student attrition. Utilizing a dataset encompassing 13 years of enrollment data from a sizable US academic institution, we developed predictive models using deep neural networks to identify students at risk of leaving school with an overall accuracy of 86%. SHapley Additive exPlanations (SHAP) was then used to enhance the transparency of the model by providing granular insights into the contribution of various factors to individual students’ dropout risks. Notably, we employed Generative AI to translate SHAP scores into comprehensible and actionable intervention recommendations presented via an interactive decision-support dashboard. This dashboard aids school administrators in designing personalized support strategies for at-risk students at the individual level. By leveraging the cutting-edge capabilities of Generative AI, our framework offers a novel approach to understanding and simulating human behavior in educational settings, emphasizing the importance of precise individual-level insights. This study bridges critical gaps in the literature by operationalizing advanced ML models and explainable AI tools to support decision-making processes in educational settings, ultimately aiming to enhance student success and institutional stability.