Time Series Foundation Models are Flow Predictors

2025-07-01

AI Summary

This study investigates the zero-shot generalization capability of time-series foundation models (TSFMs) for crowd flow forecasting, specifically under a setting that relies solely on origin-destination (OD) flow temporal evolution data—without explicit spatial information or graph-based priors. We propose an end-to-end forecasting framework built upon Moirai and TimesFM, and rigorously evaluate it across three real-world datasets. Our key contribution is the first empirical demonstration that purely temporal foundation models—trained without any spatial inductive bias—can significantly outperform conventional spatiotemporal models in zero-shot cross-domain transfer. Experimental results show average improvements of 33% in RMSE, 39% in MAE, and 49% in CPC over state-of-the-art methods. These findings validate the strong temporal modeling capacity and cross-dataset generalizability of TSFMs for crowd flow prediction, establishing a new lightweight and scalable paradigm for urban mobility forecasting.

Technology Category

• Machine Learning: Time-Series/Data Streams
• Planning, Routing, and Scheduling: Optimization of Spatio-temporal Systems
• Data Mining & Knowledge Management: Mining of Spatial, Temporal or Spatio-Temporal Data

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Web Mining and Content Analysis: Large pretrained models with web data
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications

Abstract

We investigate the effectiveness of time series foundation models (TSFMs) for crowd flow prediction, focusing on Moirai and TimesFM. Evaluated on three real-world mobility datasets-Bike NYC, Taxi Beijing, and Spanish national OD flows-these models are deployed in a strict zero-shot setting, using only the temporal evolution of each OD flow and no explicit spatial information. Moirai and TimesFM outperform both statistical and deep learning baselines, achieving up to 33% lower RMSE, 39% lower MAE and up to 49% higher CPC compared to state-of-the-art competitors. Our results highlight the practical value of TSFMs for accurate, scalable flow prediction, even in scenarios with limited annotated data or missing spatial context.

Problem

• Evaluating TSFMs for zero-shot crowd flow prediction
• Assessing performance without spatial information
• Comparing TSFMs with statistical and deep learning baselines

Innovation

• Time series foundation models for flow prediction
• Zero-shot deployment without spatial information
• Outperform statistical and deep learning baselines