Towards a FAIR Dataset for Spanish Non-Functional Requirements

Resumo

Supervised Machine Learning algorithms (ML) have enhanced the performance of the automatic non-functional requirements (NFR) classification in the Requirements Engineering domain. However, the lack of public datasets, dealing with imbalanced datasets and reproducibility are current concerns in ML experiments. We conducted a quasi-experiment to generate a dataset of NFR in the Spanish Language, following the FAIR Principles. We collected 109 requirements from an open access repository of the University of A Coru˜ na, and performed a labeling process based in the categories and subcategories of the ISO/IEC 25010 quality model. Using a Fleiss’ Kappa test we obtained a substantial agreement (0.78) at the category level and a moderate agreement (0.48) when the classification is per subcategory supervised Machine Learning algorithms (ML) have enhanced the performance of the automatic non-functional requirements (NFR) classification in the Requirements Engineering domain. However, the lack of public datasets, dealing with imbalanced datasets and reproducibility are current concerns in ML experiments. We conducted a quasi-experiment to generate a dataset of NFR in the Spanish Language, following the FAIR Principles. We collected 109 requirements from an open access repository of the University of A Coruña, and performed a labeling process based in the categories and subcategories of the ISO/IEC 25010 quality model. Using a Fleiss’ Kappa test we obtained a substantial agreement (0.78) at the category level and a moderate agreement (0.48) when the classification is per subcategory