Identification Of Felder-Silverman Based Learning Style Through Data Driven Approach In The Context Of Game Based Learning

Abstract

ABSTRACT In the era of intelligent educational technologies, personalized learning has become a central challenge in computer science and educational data mining. This thesis presents a machine learning-driven framework for real-time identification of individual learning styles, grounded in the Felder-Silverman Learning Style Model (FSLSM). Traditional self-assessment tools, such as the Index of Learning Styles (ILS), are limited in scalability, objectivity, and adaptability to dynamic digital environments. To address these limitations, this research proposes a data-driven alternative using behavioural interaction data collected through a custom-built, web-based educational crossword puzzle game. The system architecture integrates modular front-end interfaces, backend services, and a MongoDB data store to capture gameplay data such as clue-solving order, response latency, modality preference, and hint usage. These interactions are transformed into engineered features aligned with FSLSM dimensions and processed using domain-specific Multinomial Naïve Bayes classifiers. Classifier training is conducted with stratified 2-fold cross-validation, ensuring robustness and generalization. Experimental evaluation with a diverse participant cohort demonstrates particularly high predictive performance in the Active/Reflective domain, with F1 and precision scores exceeding 90%. Classifiers for the Sensing/Intuitive, Sequential/Global, and Visual/Verbal dimensions also performed strongly, achieving average F1-scores above 85%. Model predictions showed high agreement with ILS ground truth, validated through accuracy, Cohen’s Kappa, and AUC metrics. A real-time predictive web application was developed to demonstrate practical integration of behavioural analytics into adaptive learning platforms. In addition to technical validation, a structured user study revealed that a majority of participants found the game-based system more engaging, interactive, and preferable to traditional questionnaires. Feedback highlighted ease of use, improved motivation, and usability suggestions—underscoring the importance of user-centred design. These findings affirm the system’s potential not only as a robust learning style classifier but also as a learner-friendly interface. This research contributes to computer science by unifying supervised machine learning, behavioural feature engineering, and web-based system deployment into a scalable framework for adaptive learning. It also integrates HCI principles to bridge technical design with real-world usability, offering a holistic solution for intelligent learner modelling.