Structured Evaluation of Synthetic Tabular Data
This work addresses the challenge of objectively assessing synthetic data quality for researchers and practitioners dealing with incomplete or privacy-sensitive tabular data, representing an incremental improvement in evaluation methodology.
The authors tackled the lack of a coherent framework for evaluating synthetic tabular data by proposing a structured evaluation method based on distributional equivalence, showing that structurally informed synthesizers outperform others, particularly on smaller datasets.
Tabular data is common yet typically incomplete, small in volume, and access-restricted due to privacy concerns. Synthetic data generation offers potential solutions. Many metrics exist for evaluating the quality of synthetic tabular data; however, we lack an objective, coherent interpretation of the many metrics. To address this issue, we propose an evaluation framework with a single, mathematical objective that posits that the synthetic data should be drawn from the same distribution as the observed data. Through various structural decomposition of the objective, this framework allows us to reason for the first time the completeness of any set of metrics, as well as unifies existing metrics, including those that stem from fidelity considerations, downstream application, and model-based approaches. Moreover, the framework motivates model-free baselines and a new spectrum of metrics. We evaluate structurally informed synthesizers and synthesizers powered by deep learning. Using our structured framework, we show that synthetic data generators that explicitly represent tabular structure outperform other methods, especially on smaller datasets.