Causality Enhanced Origin-Destination Flow Prediction in Data-Scarce Cities DOI: 10.1145/1122445.1122456

Tao Feng, Yunke Zhang, Huandong Wang, Yong Li · 2025

Read Causality Enhanced Origin-Destination Flow Prediction in Data-Scarce Cities DOI: 10.1145/1122445.1122456 by Tao Feng, Yunke Zhang, Huandong Wang, Yong Li (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2503-06398.txt.

Introduction

Tao Feng, Yunke Zhang, Huandong Wang, and Yong Li forecast trips between zones in cities that lack rich travel data, in a paper from two thousand twenty-five. Such forecasts help planners shape roads and services, but poor cities often miss both zone traits and trip counts. The method first learns cause links among city traits in data-rich cities.

Those cause graphs then guide a rebuild of the missing traits in the poor-data city. A teaching step moves the forecast model from the rich city to the poor one with the rebuilt traits. Tests on two pairs of real city data beat leading methods and cut error by up to about a tenth. Readers get a transfer plan for data-poor planning.

Catalog record

CREATORTao Feng, Yunke Zhang, Huandong Wang, Yong Li
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE50 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2503-06398

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Causality Enhanced Origin-Destination Flow Prediction in Data-Scarce Cities DOI: 10.1145/1122445.1122456 Tao Feng, Yunke Zhang, Huandong Wang, Yong Li Abstract . Accurate origin-destination (OD) flow prediction is of great importance to developing cities, as it can contribute to optimize urban structures and layouts. However, with the common issues of missing regional features and lacking OD flow data, it is quite daunting to predict OD flow in developing cities. To address this challenge, we propose a novel Causality-Enhanced OD Flow Prediction (CE-OFP), a unified framework that aims to transfer urban knowledge between cities and achieve accuracy improvements in OD flow predictions across data-scarce cities. In specific, we propose a novel reinforcement learning model to discover universal causalities among urban features in data-rich cities and build corresponding causal graphs. Then, we further build Causality-Enhanced Variational Auto-Encoder (CE-VAE) to incorporate causal graphs for effective feature reconstruction in data-scarce cities. Finally, with the reconstructed features, we devise a knowledge distillation method with a graph attention network to migrate the OD prediction model from data-rich cities to data-scare cities. Extensive experiments on two pairs of real-world datasets validate that the proposed CE-OFP remarkably outperforms state-of-the-art baselines, which can reduce the RMSE of OD flow prediction for data-scarce cities by up to 11%. 1. Introduction Origin-Destination (OD) flow refers to the amount of population flow between two specific regions ( i.e. , the origin region and the destination region) in a city ( Lenormand et al. 2016 ; Rossi-Hansberg and Wright 2007 ) . As it reflects complex interactions between the urban structure and the travel demand of people, OD flow prediction has been widely recognized as a key enabler for a range of urban transportation applications, including transport facilities planning, traffic control and taxi dispatching ( Castiglione et al. 2015 ; Deng et al. 2016 ) . For developing cities, OD flow prediction plays an important role in optimizing the structure and layout of urban regions. Nevertheless, due the to common issues of lacking OD flow data and missing regional features, data scarcity has always been a major stumbling block in predicting accurate OD flow in developing cities. In specific, with the limited number of road sensors and traffic cameras, there is a high chance of missing features and even flow data in particular regions. Numerous existing works ( Robinson and Dilkina 2018 ; Pourebrahim et al. 2019 ; Liu et al. 2020 ) have investigated the OD flow prediction problem, but they failed to address the data scarcity problem with careful considerations. Generally, conventional methods have constrained capabilities in urban OD flow prediction, as most of them directly follow typical physical laws rather than combining flow data with features of different regions. With the rapid development of Machine Learning (ML) techniques, emerging ML models for OD flows exhibit stronger predicting capabilities, such as decision trees ( Robinson and Dilkina 2018 ; Pourebrahim et al. 2019 ) and graph neural networks ( Veličković et al. 2017 ; Liu et al. 2020 ) . Nonetheless, these approaches require a large amount of OD flow data to fit the large number of model parameters, which prevents their applications in data-scarce cities. The rising paradigm of transfer learning provides us with a promising solution to this problem ( Pan and Yang 2010 ) . Based on the transfer learning techniques, we can utilize the massive OD flow data available at data-rich cities to help us predict the OD flow of data-scarce cities. Under the circumstances, the missing regional features at the data-scarce cities become the bottleneck of the OD flow prediction. In order to solve this problem, in this paper, we focus on predicting the OD flow in data-scarce cities by first reconstructing missing urban regional features based on the observed urban regional features and then implementing OD flow prediction based on them. Specifically, both the feature reconstruction model and the OD flow prediction model are learned by utilizing the massive data in data-rich cities. However, there are several main challenges in developing such models. First, reconstructing missing features requires modeling the relationships between observed features and missing features. However, such relationships obtained from the data-rich cities are hard to generalize to the data-scarce cities due to the differences in cities. Therefore, universal relationships shared by both of data-rich cities and data-scarce cities between observed features and missing features are needed to address this problem.Second, due to the differences in the distribution and scale of urban regional features between data-rich cities and data-scarce cities, the feature reconstruction model obtained in data-rich cities should not only consider the accuracy but also its uncertainty of the reconstructed features. Traditional reconstruction methods like auto-encoder (AE) ( Semwal et al. 2017 ; Ng et al. 2011 ) model can reconstruct the features with high accuracy, but they fail to considering the uncertainty of the features caused by the differences in the distributions and scales, thus weakening its ability to generalize to different scenarios. To solve these challenges, we propose a novel a novel Causality-Enhanced OD Flow Prediction (CE-OFP) model based on a Causality Enhanced Variational Auto-Encoder (CE-VAE) to transfer urban knowledge of missing features and OD flow from data-rich cities to the data-scarce cities. For the first challenge, we propose to search the causal graph between urban features in data-rich cities as universal relationships that can be shared across multiple cities, which is recognized by many existing researches ( Wang et al. 2021 ; Zhu et al. 2019 ) . Specifically, we introduces a state-of-art reinforcement learning (RL) based method to model the causal graph searching as a sequential decision making problem, which alleviates the difficulty of NP-hard search process. The obtained causal graph depicts the essential universal relationships between urban regional features, which serves as basis for the generalization of feature reconstruction models in different cities. As for the second challenge, we propose a Causality Enhanced Variational Auto-Encoder (CE-VAE) to model the accuracy and uncertainty of missing features. CE-VAE first exploits the universal causal graph of urban regional features to find the correlation paths between observed and missing features in data-scarce cities, which guarantees the generalization ability of the feature reconstruction model across cities. It further incorporates the causal graph into the inference process of Variational Auto-Encoder (VAE) and models the accuracy and uncertainty of the missing features by outputting the mean value and the variance of the missing features, respectively. Based on the the mean value and the variance of the missing features, we utilize a GAT-based (graph attention neural network) knowledge distillation method to migrate the OD prediction model of data-rich cities to data-scarce cities, thereby enhancing the prediction performance of data-scarce cities. The contributions of our work can be summarised as follows: • We propose a novel Causality-Enhanced OD Flow Prediction (CE-OFP) model based on a Causality Enhanced Variational Auto-Encoder (CE-VAE) to transfer urban knowledge of missing features and OD flow from data-rich cities to the data-scarce cities, which enhances the prediction performance of data-scarce cities. • We propose a Causality Enhanced Variational Auto-Encoder (CE-VAE) feature reconstruction model by organically incorporating urban causal graph into the inference process of Variational Auto-Encoder to construct a latent representation vector considering the accuracy and uncertainty of the unobserved features, which promotes its

Read the complete text