Explainable Artificial Intelligence Academic Advisor for Early Academic Risk Prediction and Personalized Intervention in Higher Education

Tóm tắt

Abstract — University freshmen often face academic difficulties that may lead to course failure or dropout. This paper introduces the Artificial Intelligence Academic Advisor, an explainable machine-learning framework that combines supervised classifiers with model-agnostic interpretability to forecast early learning risk and provide individualized intervention suggestions. Using a domain-informed simulated dataset of 500 student records, we train and compare logistic regression and random forest classifiers, evaluate them with stratified five-fold cross-validation and an independent hold-out test set, and quantify uncertainty via bootstrap confidence intervals. Explanations are produced with Shapley-value attribution to identify cohort-level and case-level risk drivers. The random forest attains the best performance (cross-validated accuracy 0.92 ± 0.03; mean area under the receiver-operating-characteristic curve 0.934 ± 0.02) while explanations consistently highlight cumulative grade point average, attendance rate, and prior failed courses as primary drivers. We discuss ethical safeguards, reproducibility, limitations of simulation-based evaluation, and pathways to pilot deployment as a business-to-business educational-technology service that augments human advisors in timely and equitable interventions.

Index Terms — academic advising, early risk detection, explainable artificial intelligence, student retention.