AI-Driven Software Testing for a Test Center of Excellence: A Systematic Mapping and Evidence Synthesis

Artificial intelligence (AI) is increasingly used to automate and optimize software testing, but the evidence is distributed across heterogeneous testing activities, AI techniques, and evaluation measures. This study maps AI-driven software-testing research and translates the evidence into capabilities relevant to a Test Center of Excellence (TCoE). A systematic evidence workflow was applied to a unified audit corpus of 2,072 records, followed by focused eligibility reconciliation and full-text verification. The…

By the Truvace deskOctober 20, 20261 min read1 sourcePeer-reviewedTRV-2026-2012· cite & verify

In brief

Researchers screened a unified audit corpus of 2,072 records and narrowed it to a focused set of 29 records, of which 15 were primary empirical studies and 9 were secondary studies. The studies covered machine learning, reinforcement learning, search-based methods, deep learning, explainable AI and large language models applied to testing tasks.

The work matters because organizations running test centers of excellence face scattered evidence about which AI methods suit which testing tasks. The synthesis offers a structure that ties techniques to outcomes and oversight. Its practical value depends on resolving open gaps in reliability, reproducibility and standardized evaluation.

Main points

The final focused evidence workset contained 29 records: 15 primary empirical studies and 9 secondary studies.
Excluded records included 1 scope exclusion and 4 full-text or access exclusions.
Supported testing activities included test generation, selection and prioritization, minimization, fuzzing, validation, adaptive automation and maintenance.
The capability model includes governance with human oversight.

The gain

The evidence synthesis links AI techniques to testing activities, measurable outcomes and reusable capabilities, giving test centers a governed capability model rather than a single automation tool.

The rundown

The study applied a systematic evidence workflow to a unified audit corpus of 2,072 records, followed by eligibility reconciliation and full-text verification before arriving at the focused set.

The primary evidence spans several AI families, including evolutionary and search-based methods, deep learning, explainable AI and large language models, and maps them onto concrete testing activities.

The authors identify gaps in context awareness, test-oracle support, long-term maintenance, CI/CD integration and reproducibility that limit how far the evidence can guide adoption.