Misconception Propagation, Clustering, and Score Consistency in Acid-Base Submicroscopic Representations: A Bayesian Network and Machine Learning Approach

Abstract

Misconceptions in chemistry, particularly in submicroscopic representations such as particle structure, solution properties, ionic interaction, and chemical reactions, continue to pose challenges for students' conceptual understanding. This study aims to investigate the propagation of misconceptions, cluster students based on their misconception profiles, and analyze the consistency of scores in relation to these patterns. The participants were 52 second-semester pre-service chemistry teachers who completed a diagnostic test consisting of four particle-level diagrams with open-ended questions. Bayesian Network analysis and Granger Causality testing were employed to examine probabilistic and causal relationships between misconceptions. Clustering analysis using K-Means and visualization through t-SNE identified three distinct student groups with varying misconception levels. Score consistency analysis using correlation, ANOVA, and regression revealed that misconceptions in particle structure strongly influenced errors in other concepts and were significantly correlated with lower scores (r = -0.26). Sankey diagrams demonstrated how misconceptions in early questions propagated to subsequent concepts, indicating error flow. The findings suggest that early identification and correction of key misconceptions are crucial, and clustering analysis can inform adaptive teaching strategies. This research highlights the importance of integrating causal analysis and machine learning in chemistry education research to better understand and address patterns of student misunderstanding.