Machine Learning-Driven Personalised Learning Pathway Recommendation for Transferable Skill Enhancement in Undergraduate Academic Performance

Abstract

This study developed and evaluated a machine learning-driven personalised learning pathway recommendation system to support transferable skill development and academic performance among undergraduate students. The study used secondary data from the Open University Learning Analytics Dataset, comprising 24,998 student records after preprocessing and integration. Eighteen features were selected from the available dataset based on educational relevance and correlation analysis. K-means clustering was applied to identify four learner profiles, and the resulting cluster labels were incorporated into subsequent model development. Five classification models were trained and compared: Decision Tree, Support Vector Machine, K-Nearest Neighbours, Random Forest and XGBoost. A Stacking Hybrid model combining XGBoost, Random Forest and K-Nearest Neighbours as base learners, with Logistic Regression as the meta-learner, was then evaluated for multi-class prediction of student performance categories. The final model was tested on 5,000 student records and achieved an accuracy of 85.00%, weighted precision of 85.10%, weighted recall of 85.00% and weighted F1-score of 84.79%. Ten-fold stratified cross-validation produced a mean accuracy of 84.28%, with a standard deviation of 0.009. The recommendation component assessed nine transferable skill areas and generated personalised learning pathways using 30 curated resources across three difficulty levels. The Stacking Hybrid model correctly identified 2,150 at-risk students out of 2,405 in the 5,000-student test set, achieving an 89.40% detection rate and identifying 1,013 additional students beyond the generic baseline. The recommendation engine was separately evaluated on 500 student records and attained 92.20% accuracy based on category-to-resource matching criteria, indicating the system's effectiveness in delivering targeted transferable skill interventions. These findings indicate that machine learning-based personalisation can support more targeted academic interventions, although further validation with broader institutional datasets and longitudinal outcome measures is required.