AI TEST AUTOMATION

Abstract: Modern test automation workflows are often fragmented across isolated scripts, CI logs, repository tools, and manual triage processes, making it difficult for teams to prioritize unstable tests, interpret recurring failures, and maintain brittle UI selectors efficiently. This project presents a local-first intelligent test automation platform that unifies test management, execution tracking, failure analysis, self-healing support, and governed repository delivery within a single system built with FASTAPI, Next.js, and PostgreSQL. The platform combines a background execution queue with persisted test and run history, a deployed Random Forest-based prioritization model for ranking failure-prone tests, a TF- IDF plus K-Means clustering pipeline for grouping similar failure patterns, an explainable per-test insight layer, heuristic selector-repair suggestions with confidence thresholds, and GitHub delivery workflows that preserve human approval before branch or pull request creation. Using a 45,000-row CI/CD failure-log dataset packaged with the project, the prioritization pipeline outperformed its heuristic fallback in offline evaluation, while the clustering pipeline produced operationally interpretable but only weakly separated failure groups. The prototype also demonstrates practical explainability through surfaced model factors, cluster keywords, review states, and approval histories. However, the current implementation remains limited by simplified or synthetic data characteristics, heuristic healing logic, and partial reliance on simulated execution paths. Overall, the project shows that multiple intelligent QA functions can be integrated into one transparent, low-cost, and locally reproducible test operations platform.

Keywords: Test Automation, Continuous Integration, Test Case Prioritization, Machine Learning in Testing, Predictive Modeling, Unsupervised Learning, Test Case Prioritization, Failure Clustering, Local-First Architecture.