[Paper Review] Modeling Ride-Sourcing Matching and Pickup Processes based on Additive Gaussian Process Models

Zheng Zhu, Meng Xu+3 more|arXiv (Cornell University)|Apr 29, 2022

Transportation and Mobility Innovations4 citations

TL;DR

This paper proposes an additive Gaussian process model (AGPM) to accurately model ride-sourcing matching and pickup processes using spatial, temporal, demand, and supply covariates. Evaluated on Hangzhou data, AGPM outperforms analytical and machine learning models in estimation accuracy and enables effective idle vehicle relocation strategy design.

ABSTRACT

Matching and pickup processes are core features of ride-sourcing services. Previous studies have adopted abundant analytical models to depict the two processes and obtain operational insights; while the goodness of fit between models and data was dismissed. To simultaneously consider the fitness between models and data and analytically tractable formations, we propose a data-driven approach based on the additive Gaussian Process Model (AGPM) for ride-sourcing market modeling. The framework is tested based on real-world data collected in Hangzhou, China. We fit analytical models, machine learning models, and AGPMs, in which the number of matches or pickups are used as outputs and spatial, temporal, demand, and supply covariates are utilized as inputs. The results demonstrate the advantages of AGPMs in recovering the two processes in terms of estimation accuracy. Furthermore, we illustrate the modeling power of AGPM by utilizing the trained model to design and estimate idle vehicle relocation strategies.

Motivation & Objective

• To develop a data-driven modeling framework that balances model fit and analytical tractability for ride-sourcing operations.
• To address the gap in previous studies where model-data fitness was often neglected despite analytical tractability.
• To enable operational insights through accurate modeling of matching and pickup processes in real-world ride-sourcing systems.
• To evaluate the AGPM’s performance against analytical and machine learning models using real-world Hangzhou data.
• To demonstrate the practical utility of the trained model in designing and estimating idle vehicle relocation strategies.

Proposed method

• The AGPM framework models the number of matches or pickups as outputs, with spatial, temporal, demand, and supply variables as inputs.
• The additive structure of the Gaussian process allows for interpretable, non-linear relationships between covariates and outcomes.
• The model uses a kernel-based approach to capture complex dependencies while maintaining computational efficiency.
• Model training is performed using real-world ride-sourcing data collected in Hangzhou, China.
• The framework supports uncertainty quantification and provides probabilistic predictions for operational planning.
• The trained AGPM is applied to simulate and estimate the effectiveness of idle vehicle relocation strategies.

Experimental results

Research questions

• RQ1 How well can an additive Gaussian process model capture the dynamics of ride-sourcing matching and pickup processes compared to traditional models?
• RQ2 What is the relative estimation accuracy of AGPM versus analytical and machine learning models in predicting matches and pickups?
• RQ3 Can the AGPM framework support the design and evaluation of idle vehicle relocation strategies in ride-sourcing systems?
• RQ4 How do spatial, temporal, demand, and supply covariates influence the matching and pickup processes according to the model?
• RQ5 To what extent does the AGPM maintain analytical tractability while achieving high data fit?

Key findings

• The AGPM significantly outperforms both analytical and machine learning models in estimating the number of matches and pickups.
• The model achieves higher estimation accuracy due to its ability to capture non-linear, complex relationships between covariates and outcomes.
• The additive structure of the AGPM enables interpretable decomposition of the contribution of each covariate to the prediction.
• The trained AGPM successfully supports the design and evaluation of idle vehicle relocation strategies, demonstrating practical operational value.
• The framework maintains analytical tractability while achieving superior data fit, addressing a key limitation in prior modeling approaches.
• Empirical results on Hangzhou data confirm the robustness and generalizability of the AGPM across different spatial and temporal conditions.