TDGT Toolkit Generates High-Fidelity, Privacy-Preserving Tabular Data

The increasing demand for privacy-preserving data sharing in AI workflows has led to the development of TDGT, a new Tabular Data Generation Toolkit. This web-based solution offers an integrated platform for creating synthetic tabular data and thoroughly assessing its fidelity. It addresses common limitations in existing tools by incorporating adaptive generation strategies and comprehensive multi-metric evaluation. TDGT introduces the Adaptive Bayesian Mixture Synthesizer (ABMS), an innovative algorithm that automatically determines the optimal number of mixture components, eliminating manual hyperparameter tuning. For more complex, nonlinear data distributions, it offers VAE-ABMS, a hybrid architecture combining Variational Autoencoders with adaptive Bayesian mixture synthesis. The toolkit also includes a GPU-accelerated variant for large datasets and evaluates synthetic data using eleven statistical fidelity metrics, alongside privacy risk indicators like k-anonymity.