Modeling Population Human Mobility with Dynamic Mode Decomposition

TL;DR

This paper applies dynamic mode decomposition to model and predict population human mobility using US visitor flow data, aiming to improve understanding and management of human movement patterns.

Contribution

It introduces the use of dynamic mode decomposition for modeling population mobility and evaluates its effectiveness with real-world visitor flow data.

Findings

DMD models effectively capture mobility patterns.

Models show promising predictive accuracy.

Different low-rank structures impact model performance.

Abstract

Human mobility research concerns spatiotemporal individual and population movement. Accurate modeling and prediction of human mobility can provide opportunities to monitor, manage and optimize human movement for improved social-economic benefit. In this paper, we adopt the dynamic mode decomposition algorithm to model population human mobility using visitor flow data between different states in the United States from 2019 to 2021 [1]. We train multiple DMD models with different low rank structures, and evaluate their modeling accuracy and predictability on novel testing data.