The challenge of generating and evolving real-life like synthetic test data without accessing real-world raw data -- a Systematic Review

Researchers and computer scientists published a comprehensive systematic review on the arXiv preprint server in February 2025 addressing the critical challenge of generating realistic synthetic test data for e-Government and high-security sectors without accessing sensitive raw information. The study examines how developers can create and evolve simulated datasets that mirror the complexity of real-world scenarios while strictly adhering to privacy regulations in fields such as international information exchange, medicine, and banking. By synthesizing current industry practices, the authors seek to bridge the gap between the need for robust system testing and the legal necessity of protecting personal information from exposure. The research highlights a growing tension in software engineering where high-level system testing requires data that reflects the nuances of actual human behavior and administrative processes. Traditional methods of anonymization are often insufficient or logistically impossible due to strict data residency and privacy laws. Consequently, organizations are increasingly turning to synthetic data generation—a process that creates artificial datasets from scratch based on mathematical models or limited metadata—to ensure that applications can be thoroughly vetted for reliability without risking a breach of actual citizen or patient records. Beyond simple data creation, the review delves into the 'evolution' of synthetic data, which refers to the ability of test environments to adapt as real-world systems change over time. This is particularly vital for long-term projects like cross-border data sharing or digital banking platforms, where the underlying data structures frequently shift. The paper identifies key methodologies and tools currently utilized by practitioners to maintain the 'real-life-like' quality of these datasets, ensuring that test results remain valid even as the simulated environments grow in scale and complexity.