ChiMMD: a Chicago Multi-Modal Dataset for Socio-economic and Urban Analysis

Socio-economic data plays an important role in understanding how different aspects of a city—such as traffic, housing, social activity—interact. While many unimodal datasets exist for major cities, multi-modal datasets remain limited. In this work, we introduce ChiMMD, a large-scale multi-modal dataset for the city of Chicago that integrates traffic, real estate, points of interest, and social media activity into a unified framework. We describe the data processing pipeline that allows heterogeneous data sources to be mapped to a common set of traffic zones. We establish traffic forecasting as the primary benchmark and evaluate a state-of-theart spatio-temporal model on ChiMMD, comparing unimodal and multi-modal variants. We analyze how auxiliary modalities affect forecasting performance across different horizons. Our code is available at https://github.com/donaldlin30/ChiMMD-benchmark