Combining Clustering and Association Rule Mining in Educational Data: Patterns in Diagnostic Assessments

Resumo

This study explores the application of pattern mining and unsupervised learning techniques to analyze student performance data from large-scale educational assessments in Brazilian public schools. The dataset comprises detailed binary records of learning descriptors associated with students from the final year of secondary education. The proposed approach begins with the FP-Growth algorithm to extract initial association rules from the full dataset. These rules are then transformed into numerical vectors and clustered using HDBSCAN to identify student groups based on rule similarity. FP-Growth is subsequently reapplied within each cluster to generate localized association rules, enhancing interpretability and pedagogical relevance. Results suggest that this sequential strategy improves the contextual coherence and clarity of the discovered patterns when compared to global rule extraction. This work contributes to the field of Educational Data Mining by demonstrating how combining rule mining and clustering in multiple stages can reveal differentiated learning profiles and support more targeted educational interventions.