Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Deep Contrastive Learning for Feature Alignment: Insights from Housing-Household Relationship Inference

Xiao Qian, Shangjia Dong, Rachel Davidson

TL;DR

This work tackles the problem of learning joint housing-household relationships from tabular ACS PUMS microdata in the absence of explicit ground-truth labels. It introduces a dual-encoder deep contrastive learning framework that leverages co-occurrence as a self-supervised pretext task, augmented by Bisecting K-Means clustering and a sigmoid loss with momentum distillation to handle noise and many-to-many matching. Synthetic-ground-truth experiments show substantial improvements over the state-of-the-art SubTab baseline, and Delaware real data confirms high precision and ranking quality, with North Carolina demonstrating robust transferability. Post-hoc SHAP analysis reveals tenure and mortgage information as key drivers, offering interpretable insights for housing policy and disaster resilience planning.

Abstract

Housing and household characteristics are key determinants of social and economic well-being, yet our understanding of their interrelationships remains limited. This study addresses this knowledge gap by developing a deep contrastive learning (DCL) model to infer housing-household relationships using the American Community Survey (ACS) Public Use Microdata Sample (PUMS). More broadly, the proposed model is suitable for a class of problems where the goal is to learn joint relationships between two distinct entities without explicitly labeled ground truth data. Our proposed dual-encoder DCL approach leverages co-occurrence patterns in PUMS and introduces a bisect K-means clustering method to overcome the absence of ground truth labels. The dual-encoder DCL architecture is designed to handle the semantic differences between housing (building) and household (people) features while mitigating noise introduced by clustering. To validate the model, we generate a synthetic ground truth dataset and conduct comprehensive evaluations. The model further demonstrates its superior performance in capturing housing-household relationships in Delaware compared to state-of-the-art methods. A transferability test in North Carolina confirms its generalizability across diverse sociodemographic and geographic contexts. Finally, the post-hoc explainable AI analysis using SHAP values reveals that tenure status and mortgage information play a more significant role in housing-household matching than traditionally emphasized factors such as the number of persons and rooms.

Deep Contrastive Learning for Feature Alignment: Insights from Housing-Household Relationship Inference

TL;DR

This work tackles the problem of learning joint housing-household relationships from tabular ACS PUMS microdata in the absence of explicit ground-truth labels. It introduces a dual-encoder deep contrastive learning framework that leverages co-occurrence as a self-supervised pretext task, augmented by Bisecting K-Means clustering and a sigmoid loss with momentum distillation to handle noise and many-to-many matching. Synthetic-ground-truth experiments show substantial improvements over the state-of-the-art SubTab baseline, and Delaware real data confirms high precision and ranking quality, with North Carolina demonstrating robust transferability. Post-hoc SHAP analysis reveals tenure and mortgage information as key drivers, offering interpretable insights for housing policy and disaster resilience planning.

Abstract

Housing and household characteristics are key determinants of social and economic well-being, yet our understanding of their interrelationships remains limited. This study addresses this knowledge gap by developing a deep contrastive learning (DCL) model to infer housing-household relationships using the American Community Survey (ACS) Public Use Microdata Sample (PUMS). More broadly, the proposed model is suitable for a class of problems where the goal is to learn joint relationships between two distinct entities without explicitly labeled ground truth data. Our proposed dual-encoder DCL approach leverages co-occurrence patterns in PUMS and introduces a bisect K-means clustering method to overcome the absence of ground truth labels. The dual-encoder DCL architecture is designed to handle the semantic differences between housing (building) and household (people) features while mitigating noise introduced by clustering. To validate the model, we generate a synthetic ground truth dataset and conduct comprehensive evaluations. The model further demonstrates its superior performance in capturing housing-household relationships in Delaware compared to state-of-the-art methods. A transferability test in North Carolina confirms its generalizability across diverse sociodemographic and geographic contexts. Finally, the post-hoc explainable AI analysis using SHAP values reveals that tenure status and mortgage information play a more significant role in housing-household matching than traditionally emphasized factors such as the number of persons and rooms.

Paper Structure

This paper contains 17 sections, 6 equations, 10 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Literature Review
  3. Methodology
  4. Housing-Household co-occurrence in microdata
  5. Bisecting K-Means clustering of households and housing units
  6. Deep contrastive learning (DCL) model architecture
  7. Performance Evaluation Metrics
  8. Performance Evaluation of DCL in a Controlled Experiment
  9. Synthetic ground truth generation
  10. Performance evaluation of joint variable relationship learning
  11. Case Study in Delaware and North Carolina
  12. Microdata for housing-household relationship modeling and testing
  13. Housing-household joining performance in Delaware
  14. Sensitivity analysis of cluster size
  15. Transferability test in North Carolina
  16. ...and 2 more sections

Figures (10)

  • Figure 1: Microdata table illustration and descriptive statistics
  • Figure 2: Illustration of the co-occurring pretext task using the ACS dataset.
  • Figure 3: Joint Housing-household relationship learning pipeline
  • Figure 4: Illustration of the motivation and process of bisecting K-Means clustering.
  • Figure 5: Impact of clustering granularity on label matching quality
  • ...and 5 more figures