A comprehensive GitHub Actions dataset for CI/CD pipeline failure forecasting

The GitHub Actions CI/CD Failure Prediction Dataset (GADFPD) is a curated research dataset that is designed to facilitate machine learning research that predicts Continuous Integration and Continuous Deployment (CI/CD) pipeline failures. The dataset is made up of software repository metrics, workflow execution details, commit metadata, and historical pipeline results from Open-Source GitHub repositories that use GitHub Actions as their CI/CD platform.
Every record in the dataset is a GitHub actions workflow, and is classified by the outcome of the workflow run. The data consists of attributes from software repositories and CI/CD pipelines, such commit characteristics, developer activity, repository statistics, workflow configuration, execution metadata, and pipeline history. These features are designed to help develop and test predictive models that can predict pipeline failures before they occur.
The dataset was created for an academic research project titled “A Machine Learning Approach for Forecasting CI/CD Pipeline Failures Using Software Repository and Pipeline Data.” It can be used for software engineering research, DevOps analytics, repository mining, machine learning, and predictive software quality research. Potential applications include CI/CD failure prediction, build health monitoring, software quality assessment, and intelligent DevOps decision support systems, among others.
The dataset is made available for research and educational purposes, and is meant to encourage reproducibility, benchmarking, and further study of data-driven methods to make modern CI/CD pipelines more reliable and efficient. It builds upon public information on GitHub Actions workflows provided by GitHub and is compatible with the growing body of studies leveraging execution histories of GitHub Actions for empirical software engineering research.