A Deep Generative Framework for Joint Households and Individuals Population Synthesis

TL;DR

This work addresses the need for synthetic populations that preserve household–individual and within-household correlations while matching census tract marginals. It presents a deep generative framework based on Variational Autoencoders (VAEs) with a novel data restructuring that encodes each household and all its members in a single record, a parameter-efficient transfer-learning pipeline to shift from state-level microdata to tract-level marginals, and a Decoupled Binary Cross Entropy (D-BCE) loss to avoid strict one-to-one replication. The method learns from a microdata matrix $X riangleq \\mathbb{R}^{N imes D}$ and targets census tract marginals, producing synthetic inventories that align with marginals while maintaining realistic household–individual and individual–individual relationships. Across Delaware and North Carolina, pre-training captures joint distributions consistent with microdata, while fine-tuning achieves tract-level marginal alignment and preserves privacy as measured by Distances to Closest Records (DCR), enabling scalable, transferable synthetic population generation for policy analysis and infrastructure planning.

Abstract

Household and individual-level sociodemographic data are essential for understanding human-infrastructure interaction and policymaking. However, the Public Use Microdata Sample (PUMS) offers only a sample at the state level, while census tract data only provides the marginal distributions of variables without correlations. Therefore, we need an accurate synthetic population dataset that maintains consistent variable correlations observed in microdata, preserves household-individual and individual-individual relationships, adheres to state-level statistics, and accurately represents the geographic distribution of the population. We propose a deep generative framework leveraging the variational autoencoder (VAE) to generate a synthetic population with the aforementioned features. The methodological contributions include (1) a new data structure for capturing household-individual and individual-individual relationships, (2) a transfer learning process with pre-training and fine-tuning steps to generate households and individuals whose aggregated distributions align with the census tract marginal distribution, and (3) decoupled binary cross-entropy (D-BCE) loss function enabling distribution shift and out-of-sample records generation. Model results for an application in Delaware, USA demonstrate the ability to ensure the realism of generated household-individual records and accurately describe population statistics at the census tract level compared to existing methods. Furthermore, testing in North Carolina, USA yielded promising results, supporting the transferability of our method.

Figures (12)

• Figure 1: Comparison of household income distribution. (a) Map of the state of Delaware with two randomly selected census tracts, A and B; and (b) marginal distributions of household income for the state (microdata), census tract A, and census tract B.
• Figure 2: The end-to-end deep generative pipeline for synthetic household-individual inventory development
• Figure 3: Data restructuring procedure illustration
• Figure 4: Parameter-efficient transfer learning procedure
• Figure 5: VAE structure in the proposed deep generative framework