OntoFS: An Ontology for Feature Selection Experiments

Feature selection is an important component of the machine learning pipeline, typically integrated in the data pre-processing and model training stages. However, feature selection itself is a complex process, and evaluating a feature selection method requires well-defined steps and structured outputs to support the evaluation of outcomes, whether it involves method comparison, stability assessment, reproducibility, or insights generation. To address the lack of standardized experimental representations, we propose the OntoFS, an ontology designed to provide a shared conceptual backbone for feature selection experiments. Developed using the Web Ontology Language (OWL), OntoFS enhances explainability and facilitates the systematic generation of insights from experimental outcomes. We provide three use cases to demonstrate its utility and applicability.