Predicting academic performance: a comparative study of UTME and Direct entry students using a hybrid Kernel model

Abstract

This Study introduces a novel Hybrid kernel Model (HKM) that integrates weighted and unweighted kernel regression within a unified framework to predict and compared the academic performance of students admitted through two unified pathways-Unified tertiary Matriculation Examination (UTME) and Direct Entry (DE) at Sokoto State University. While prior research has employed single kernel model and machine learning techniques in educational data mining, these studies often overlook admission mode differences and integrated kernel approaches, limiting their predictive accuracy and generalizability. Using a dataset spanning 2018-2025, partitioned into training (2018-2023) and testing (2024-2025) sets, the HKM was evaluated using MSE, RMSE, MAE and . Results consistently show DE student outperforming UTME students across graduation metrics, including higher class: first class and second-class upper percentages, and lower-class rate: second lower and third-class percentage. Performance score on a 4-point scale ranged from 3.09-3.28 for DE versus 2.65-2.90 for UTME. Statistical tests confirmed admission mode as a significant predictor (P < 0,05), with prediction confidence exceeding 84% for both groups. This study fills a critical gap by disaggregating student populations and offering pathway-specific insights, challenging homogeneous treatment in existing literature. Its contributions include a replicable methodological framework and actionable recommendations for differentiated admission orientation, and support systems. The HKM proves analytics in education, supporting equitable policy and institutional quality enhancement.