Ontology-driven Feature Engineering For Machine Learning

This study proposes an ontology based feature engineering methodology to enhance the performance and interpretability of machine learning models. By developing an educational ontology structured around a student performance dataset, semantic structures were integrated into the processing pipeline to conceptually group and justify data attributes. Different models were implemented and compared across three tasks: final g rade p rediction, a cademic p erformance classification, and anomaly detection, contrasting traditional approaches with ontology-enhanced versions. While statistical models outperformed the quantitative metrics, the ontology-driven models proved competitive, more structured, and offered greater traceability. This study highlights the potential of ontologies as a complementary tool for machine learning, particularly in contexts where knowledge sustainability is crucial. This project establishes a foundation for further applications in educational domains and other areas involving high semantic complexity.