Generative Artificial Intelligence-Assisted Software Testing Using Automated Test Case Generation and Defect Prediction

Abstract

Software quality assurance is challenged by rapid code changes, incomplete requirements, and the cost of writing and maintaining effective tests. This paper proposes an integrated framework combining generative artificial intelligence for context-aware unit test construction with supervised defect prediction for risk-prioritized execution. The workflow collects source code, change histories, requirements and historical fault labels; produces constrained test candidates; validates syntax and assertions; predicts defect likelihood; and schedules tests under execution budgets. A formal multi-objective objective balances coverage, fault discovery, runtime and flaky-test penalties. A reproducible evaluation protocol is defined using chronological splits, mutation testing, calibration, leakage controls and human oracle review. Because no experiment was executed for this manuscript, numerical tables and charts are explicitly labeled illustrative scenario data, not measured evidence. In the constructed demonstration, the integrated configuration yields 87% branch coverage, 76% mutation score, 0.84 defect F1 and 38 minutes execution time, compared with 65%, 49%, 0.72 and 55 minutes for the rule-based baseline. These values illustrate how a future experiment should report trade-offs rather than establishing effectiveness. The contribution is a transparent, auditable design for linking generated tests and risk forecasts, with clear limitations and a validation roadmap.