DPTVAE: Data-driven prior-based tabular variational autoencoder for credit data synthesizing

Yandan Tan, Hongbin Zhu, Jie Wu, Hongfeng Chai

Abstract: Highlights•Data-driven prior-based tabular variational autoencoder (DPTVAE) for end-to-end synthesizing credit data.•Value importance-based automatic discrete continuous feature classification.•Adaptively encoding different complexity based on Binning Gaussian probability density-Variational Gaussian Mixture (BGPD-VGM).•Conditional decoding for DPTVAE to synthesize imbalanced discrete columns.