Skip to content
Conference

Predicting First Semester Academic Performance Using Pre Enrolment Machine Learning

Sep 2026 · International Conference on Communications and Information Technology · pp. 1-6 · 0 citations · 43 references

Abstract

Early identification of academically vulnerable higher education students enables timely intervention, yet predictive models often act as opaque black boxes or depend on concurrent behavioural indicators. This study models first-semester Grade Point Average using explainable machine learning based strictly on pre-enrolment attributes. Data were collected from 5,484 first-year students across 73 features, including high-school grades, English test scores, demographics, and 42 OMNI Personality Inventory dimensions. To prevent data leakage, all concurrent behavioural metrics were excluded. Predictive models were evaluated using an 80:20 stratified split with 5-fold cross-validation, comparing Histogram-based Gradient Boosting, Ridge regression, and Random Forest against a baseline. Model interpretability was established via SHAP and permutation importance. On the test set of 1,097 students, the tuned Gradient Boosting model achieved a root mean squared error of 0.3403, a mean absolute error of 0.2392, and an R squared of 0.4773, marking a 28% error reduction over baseline. Study programme, English proficiency, high-school grades, and gender proved to be the primary individual drivers, whereas the personality dimensions collectively contributed 44% of total SHAP magnitude. Institutional administrators can reliably deploy this pre-enrolment risk screening while pairing low-performing predictions with dedicated secondary monitoring.

View source

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.