Machine Learning-Based Prediction of Software Test Failures Using Code and Execution Metrics

Resumen

Test suites in large software projects grow to thousands of cases, and running them all on every change is slow and costly, motivating predictive approaches that anticipate which tests are likely to fail. This study develops a machine-learning approach to predict software test failures using a combination of static code metrics and dynamic execution metrics, enabling smarter test selection and prioritization. We assembled a dataset linking code characteristics, historical execution outcomes, and change information to test-case results, and trained classifiers to predict failure for each test on a given build. The approach was benchmarked against baseline strategies and evaluated with both classification metrics and test-prioritization measures. The proposed gradient-boosting model achieved an F1 of 0.78 and AUC of 0.88, outperforming logistic-regression and history-only baselines....