Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

HouseTS: A Large-Scale, Multimodal Spatiotemporal U.S. Housing Dataset and Benchmark

Shengkun Wang, Yanshen Sun, Fanglan Chen, Linhan Wang, Naren Ramakrishnan, Chang-Tien Lu, Yinlin Chen

TL;DR

HouseTS addresses the need for a large-scale, open, multimodal spatiotemporal housing dataset designed for long-horizon forecasting. It introduces a ZIP-code panel covering $6{,}000$ ZIPs in $30$ metros from $2012$ to $2023$, aligning housing signals, POIs, census covariates, and time-stamped aerial imagery under a unified preprocessing protocol. The work defines standardized univariate and multivariate forecasting benchmarks across six window configurations and three horizons, evaluating 16 model families including classical, deep learning, and pretrained time-series foundations; results show strong performance of simple linear baselines when properly preprocessed, with foundation models offering robustness but not consistently exceeding traditional baselines. Additionally, HouseTS provides image-derived textual change annotations with disagreement-based reliability metadata to enable interpretable multimodal analyses and reliability-aware dataset usage, and it releases the dataset, code, and documentation for reproducible research in spatiotemporal learning.

Abstract

Accurate long-horizon house-price forecasting requires benchmarks that capture temporal dynamics together with time-varying local context. However, existing public resources remain fragmented: many datasets have limited spatial coverage, temporal depth, or multimodal alignment; the robustness of modern deep forecasters and time-series foundation models on housing data is not well characterized; and aerial imagery is rarely leveraged in a time-aware and interpretable manner at scale. To bridge these gaps, we present HouseTS (House Time Series), a multimodal spatiotemporal dataset for ZIP-code-level housing-market analysis, covering monthly signals from March 2012 to December 2023 across over 6,000 ZIP codes in 30 major U.S. metropolitan areas. HouseTS aligns monthly housing-market indicators, monthly POI dynamics, and annual census-based socioeconomic variables under a unified schema, and includes time-stamped annual aerial imagery. Building on HouseTS, we define standardized long-horizon forecasting tasks for univariate and multivariate prediction and benchmark 16 model families spanning statistical methods, classical machine learning, deep neural networks, and time-series foundation models in both zero-shot and fine-tuned modes. We also provide image-derived textual change annotations from multi-year aerial image sequences via a vision--language pipeline with LLM-as-judge and human verification to support scalable interpretability analyses. HouseTS is available on Kaggle, with code and documentation on GitHub.

HouseTS: A Large-Scale, Multimodal Spatiotemporal U.S. Housing Dataset and Benchmark

TL;DR

HouseTS addresses the need for a large-scale, open, multimodal spatiotemporal housing dataset designed for long-horizon forecasting. It introduces a ZIP-code panel covering ZIPs in metros from to , aligning housing signals, POIs, census covariates, and time-stamped aerial imagery under a unified preprocessing protocol. The work defines standardized univariate and multivariate forecasting benchmarks across six window configurations and three horizons, evaluating 16 model families including classical, deep learning, and pretrained time-series foundations; results show strong performance of simple linear baselines when properly preprocessed, with foundation models offering robustness but not consistently exceeding traditional baselines. Additionally, HouseTS provides image-derived textual change annotations with disagreement-based reliability metadata to enable interpretable multimodal analyses and reliability-aware dataset usage, and it releases the dataset, code, and documentation for reproducible research in spatiotemporal learning.

Abstract

Accurate long-horizon house-price forecasting requires benchmarks that capture temporal dynamics together with time-varying local context. However, existing public resources remain fragmented: many datasets have limited spatial coverage, temporal depth, or multimodal alignment; the robustness of modern deep forecasters and time-series foundation models on housing data is not well characterized; and aerial imagery is rarely leveraged in a time-aware and interpretable manner at scale. To bridge these gaps, we present HouseTS (House Time Series), a multimodal spatiotemporal dataset for ZIP-code-level housing-market analysis, covering monthly signals from March 2012 to December 2023 across over 6,000 ZIP codes in 30 major U.S. metropolitan areas. HouseTS aligns monthly housing-market indicators, monthly POI dynamics, and annual census-based socioeconomic variables under a unified schema, and includes time-stamped annual aerial imagery. Building on HouseTS, we define standardized long-horizon forecasting tasks for univariate and multivariate prediction and benchmark 16 model families spanning statistical methods, classical machine learning, deep neural networks, and time-series foundation models in both zero-shot and fine-tuned modes. We also provide image-derived textual change annotations from multi-year aerial image sequences via a vision--language pipeline with LLM-as-judge and human verification to support scalable interpretability analyses. HouseTS is available on Kaggle, with code and documentation on GitHub.

Paper Structure

This paper contains 15 sections, 4 equations, 9 figures, 6 tables.

Table of Contents

  1. INTRODUCTION
  2. Related Work
  3. HouseTS Dataset
  4. Benchmark
  5. Benchmark evaluation methodology
  6. Benchmark evaluation results
  7. Image-Derived Textual Annotations
  8. VLMs aerial image annotations
  9. Disagreement-based reliability with LLM-as-a-judge and human verification
  10. Conclusion
  11. Feature Analysis
  12. Exploratory PCA analysis
  13. Geographic Graph Neural Network Baselines
  14. Ethics, Fairness, and Data Licensing
  15. Aerial Image and Prompts for Image-Derived Textual Annotations

Figures (9)

  • Figure 1: HouseTS overview: 30 U.S. metropolitan areas and aligned ZIP-level modalities.
  • Figure 2: Distribution of house prices before (left) and after log transformation (right).
  • Figure 3: VLM-based semantic annotation pipeline.
  • Figure 4: Model-specific distributions of self-reported uncertainty scores (1--5).
  • Figure 5: Deviation of each annotator relative to the multi-model consensus (median), measured as mean absolute difference over discrete score fields. Lower deviation indicates closer alignment to consensus.
  • ...and 4 more figures