Students’ Learning Style in An Educational Gaming System: A Twin-Level Prediction Model

Considering how students learn with educational technologies is a necessity in improving learning outcomes in personalized and adaptive learning systems. This research looks into the implementation of a twin predictive model to categorize and predict students’ learning styles within the context of a serious educational game. In particular, the study exploits the Gregory Learning Style Model consisting of Concrete Sequential, Abstract Sequential, Abstract Random, and Concrete Random learners to identify cognitive preference. This study used interaction data from Jo Wilder and the Capitol Case educational game alongside a hybrid machine learning model which employed K-Means Clustering for unsupervised classification and behaved as a Decision Tree Classifier for supervised classification. Behavioural features that included session activity, game progress, and time spent enabled the model to accurately categorize learners into correct Gregory learning style groups. Assessment of the model’s performance showed high precision, with the decision tree classifier attaining 92.4% accuracy and over 90% F1-score in the majority of learning style classes. The analysis showed also that the total time spent in the game 49.60% and session level 26.77% were the strongest predictors of students’ learning styles. This study helps advance educational data mining, adaptive learning, and serious game design by showing an accurate model for learning style prediction. These findings are critical for personalized instructional design, individualized real-time support, and the integration of behaviour analytics into instructional systems. This work also supports the automated learner model design and the adaptive content delivery within digital learning environments.