AIODC-Based Classification of 42 Keras Bugs from an Open Benchmark Dataset

This repository offers a meticulously organized and refined dataset comprising 42 Keras defects sourced from GitHub, as detailed in the benchmark study: Morovati, M.M., Nikanjam, A., Khomh, F., & Jiang, Z.M. (2023). Bugs in Machine Learning-Based Systems: A Faultload Benchmark, Empirical Software Engineering, 28(3), 62. The benchmark highlights 42 Keras bugs that have been identified from GitHub. This dataset faithfully reproduces those defects as they were originally published and incorporates a comprehensive AIODC classification, which includes: AI Attribute Severity Attribute Impact Attribute utilizing the AI and AIP Quality Models (based on Kharchenko et al., Sensors, 2022). The repository includes: dataset.csv – structured defect data accompanied by AIODC classifications AIODC_Classification_Guidelines.pdf – an explanation of the methodology employed README.md – documentation and notes on usage. This dataset facilitates reproducible research in the analysis of AI defects, evaluation of AI quality, and empirical investigations into faults within ML systems. All data is made available under the CC BY 4.0 license.