Self-Healing Test Automation Dataset: 11,100 Experiment Rows with DOM-Similarity ML Models

A research dataset supporting an empirical study on AI-driven self-healing test automation. Contains 11,100 experiment rows from nine Selenium test suites across five progressively mutated UI versions under five ranking configurations (heuristic, ML-Gradient Boosting, ML-XGBoost, hybrid-GB, hybrid-XGB). Key contents: 2,400+ broken locator events per mode, 13,000+ per-candidate feature records with 13 DOM similarity features, trained Gradient Boosting (AUC=0.9993) and XGBoost (AUC=0.9994) classifiers, and full experiment CSVs. The hybrid ensemble achieves 68.63% healing success rate, a statistically significant +4.05 pp improvement over the heuristic baseline. Both algorithms converge to identical hybrid HSR, demonstrating algorithm-agnostic robustness.