A multi-project, multi-domain, and feature-engineered Itemlet dataset for empirical software engineering
TL;DR
The Itemlet dataset is created, which is compliant with agile-based software engineering tasks and demonstrates that the dataset complies with FAIR principles on fifteen criteria; except for one, all criteria are completely satisfied.
Abstract
The ability to benchmark new approaches and replicate already published research in empirical software engineering has suffered from limited advancement over time because there was no large-scale, multi-domain dataset containing both raw workflow data and derived features in one place. To address this, we created the Itemlet dataset, which is compliant with agile-based software engineering tasks. It is comprised of 727,282 data points from 204 projects with publicly accessible Jira issue trackers across 19 different domains. There are 108 features per sample, which consist of 60 raw structural fields retrieved from the JIRA REST API and 48 derived features. The raw structural fields and the derived features were designed to support three types of research tasks: (1) sprint planning and effort proxy analysis; (2) requirements prioritization; and (3) complexity classification. The effort-related fields are proxies reflecting how agile teams record effort; the dataset does not contain independently validated deliverable size measures (such as function points) required for formal effort estimation in the classical sense. We also demonstrated that the dataset complies with FAIR principles on fifteen criteria; except for one, all criteria are completely satisfied. The dataset can be accessed under CC-BY 4.0 license terms and has been archived at Zenodo with a DOI number: https://zenodo.org/records/19411554 .