What the study found
The study found that the proposed Hierarchical Feature Generation Framework (HFGF) improved how well synthetic tabular data preserved functional dependencies and logical dependencies. The authors report that this was seen across multiple generative models and datasets.
Why the authors say this matters
The authors say this matters because synthetic tabular data is increasingly used in privacy-sensitive domains such as healthcare, where preserving relationships between features is important. The findings indicate that better retention of these dependencies can improve the structural fidelity and utility of synthetic data.
What the researchers tested
The researchers proposed HFGF, a framework that first generates independent features with a standard generative model and then reconstructs dependent features using predefined functional dependency (FD) and logical dependency (LD) rules. They evaluated it on four benchmark datasets with known dependencies and three publicly available real-world datasets, using six generative models including CTGAN, TVAE, and GReaT.
What worked and what didn't
The reported results show that HFGF improved preservation of FDs and LDs across the tested generative models. The abstract also says utility analysis and qualitative dependency visualizations further showed significant improvements in structural fidelity and utility of the synthetic tabular data.
What to keep in mind
The available summary does not describe detailed limitations, statistical values, or failure cases. The evaluation was based on the datasets and models named in the abstract, so the claims are limited to those tested settings.
Key points
- HFGF is a framework for synthetic tabular data generation that handles dependent features after generating independent ones.
- The study reports improved preservation of functional dependencies and logical dependencies.
- The framework was tested on four benchmark datasets and three real-world datasets.
- Six generative models were used, including CTGAN, TVAE, and GReaT.
- The abstract says structural fidelity and utility were significantly enhanced.
Disclosure
- Research title:
- Dependency-aware synthetic tabular data improves relationship preservation
- Authors:
- Chaithra Umesh, Kristian Schultz, Manjunath Mahendra, Saptarshi Bej, Olaf Wolkenhauer
- Publication date:
- 2026-04-22
- OpenAlex record:
- View
Get the weekly research newsletter
Stay current with scholarly research without reading academic papers — one filtered digest, every Friday.