An Integrated Machine Learning Framework for Early Prediction of Student Academic Performance in University Management Systems

Forecasting student academic standing at an early stage is a critical challenge in higher education. This paper presents a supervised machine learning classification framework applied to final examination records from 1,015 students across the English and Translation Departments of Nawroz University, Duhok, Kurdistan Region, Iraq, spanning Semesters 1–6 under the Bologna Process. Six classifiers were evaluated using stratified five-fold crossvalidation and an independent held-out test set. Each student’s six course final marks served as input features; the output was one of four GPA-derived tiers (Excellent, Good, Satisfactory, Poor/At-Risk). SVM (RBF) achieved the strongest performance: 96.06% test accuracy, 94.09% cross-validated accuracy, and 92.55% macro F1-score. Results show that a lightweight ML pipeline using only routine assessment data can exceed 94% prediction accuracy, making it a viable early-warning component within existing

University Management System infrastructure.