CI/CD Build Success Prediction Data from Travis and GitHub Actions

175,706 automated software build records from two open-source CI/CD platforms, TravisTorrent and GHALogs, spanning 10 years from 2013 to 2023. The dataset supports research on predicting build success while avoiding temporal data leakage, a methodological flaw identified by a three-type taxonomy. It was created by Lalit Narayan Mishra and is licensed under CC-BY-4.0.

Use Cases

• Training models to predict CI/CD build success based on static project metadata.
• Studying the impact of project maturity and build history on build outcomes.
• Benchmarking methods for detecting and preventing temporal data leakage in software prediction tasks.
• Analyzing differences in predictive features between older (Travis) and modern (GitHub Actions) CI/CD platforms.

Strengths

• Contains 175,706 build records from two distinct platforms, providing a substantial sample size.
• Covers a 10-year time range (2013-2023), allowing for longitudinal analysis.
• Includes results from a validated methodology for preventing temporal data leakage, a key methodological contribution.
• Models built on the data achieved 82.73% and 83.30% accuracy, demonstrating practical utility.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count per platform is partially known (100k for TravisTorrent, 75,706 for GHALogs), but the exact structure of the 5.5 KB XLS file is unclear.
• Data may reflect bias inherent to the specific open-source projects monitored by the two CI/CD platforms.

Provenance

Source

TravisTorrent and GHALogs (GitHub Actions) CI/CD platforms.

Collection Method

Data collection from automated build systems of open-source projects.

Time Range

2013-2023

Freshness

Last updated 2026-05-27 17:39:32; freshness should be verified.