Improving Online Learning via VARK Learning Styles and Machine Learning-Driven Personalisation

Abstract

Understanding student engagement and academic performance is crucial in online learning environments. However, many learning management systems (LMS) lack mechanisms to adapt to diverse learning styles and support meaningful collaboration. This study addresses these challenges by proposing a Personalised and Collaborative Learning Experience (PCLE) framework that integrates the Visual, Auditory, Reading/Writing, and Kinaesthetic (VARK) learning style model with collaborative filtering techniques. Unlike existing approaches that rely only on rating data, PCLE incorporates personalised learning styles into the recommendation process to create learner-centred outcomes. To overcome the lack of publicly available datasets containing personalised learning style data, a self-collected dataset was developed to reflect authentic learner preferences. Benchmark datasets from Coursera and Udemy were also used to validate baseline collaborative filtering performance. Three machine learning models—K-Nearest Neighbours (KNN), Singular Value Decomposition (SVD), and Neural Collaborative Filtering (NCF)—were applied and evaluated using Mean Absolute Error (MAE), Hit Rate (HR), and Average Reciprocal Hit Ranking (ARHR). Results from the benchmark datasets confirmed earlier findings that KNN performs well on structured review data, while the self-collected dataset demonstrated the added value of integrating learning styles. The self-collected dataset was evaluated separately, incorporating personalised learning styles into the recommendation process. This dataset represents the main contribution of the study, as VARK preferences were embedded alongside course interactions to extend recommendations beyond standard rating-based methods. This study highlights how personalisation and collaboration can be integrated into one framework to enhance learner engagement. While the same models were applied, embedding learning preferences into the self-collected dataset represents a methodological enhancement rather than a direct comparison with existing datasets. Findings highlight the role of dataset characteristics in shaping both accuracy and ranking quality and show how the PCLE framework balances personalisation with collaboration to support learner-centred outcomes. Future research should expand dataset diversity, include additional learner attributes, and explore advanced recommendation models to further optimise adaptability and performance.