Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

AMBIT: Augmenting Mobility Baselines with Interpretable Trees

Qizhi Wang

TL;DR

AMBIT tackles the accuracy-interpretability gap in city-scale, hourly OD flow prediction by coupling a physics-informed baseline with a learned, interpretable residual. The method uses residual learning in log space, where $r = \log(1 + T)$ minus a physics baseline term, and combines this with SHAP explanations to preserve transparency while approaching the accuracy of strong tree-based predictors. A comprehensive empirical study on NYC taxi OD data shows PPML gravity as the most reliable physical baseline at hourly resolution, with AMBIT using POI-anchored or time-segment residuals delivering competitive performance and robust generalization under spatial holdouts. The work provides a reproducible pipeline, rich diagnostics, and actionable insights for urban decision-making, offering a practical path toward interpretable, scalable mobility modeling without sacrificing accuracy.

Abstract

Origin-destination (OD) flow prediction remains a core task in GIS and urban analytics, yet practical deployments face two conflicting needs: high accuracy and clear interpretability. This paper develops AMBIT, a gray-box framework that augments physical mobility baselines with interpretable tree models. We begin with a comprehensive audit of classical spatial interaction models on a year-long, hourly NYC taxi OD dataset. The audit shows that most physical models are fragile at this temporal resolution; PPML gravity is the strongest physical baseline, while constrained variants improve when calibrated on full OD margins but remain notably weaker. We then build residual learners on top of physical baselines using gradient-boosted trees and SHAP analysis, demonstrating that (i) physics-grounded residuals approach the accuracy of a strong tree-based predictor while retaining interpretable structure, and (ii) POI-anchored residuals are consistently competitive and most robust under spatial generalization. We provide a reproducible pipeline, rich diagnostics, and spatial error analysis designed for urban decision-making.

AMBIT: Augmenting Mobility Baselines with Interpretable Trees

TL;DR

AMBIT tackles the accuracy-interpretability gap in city-scale, hourly OD flow prediction by coupling a physics-informed baseline with a learned, interpretable residual. The method uses residual learning in log space, where minus a physics baseline term, and combines this with SHAP explanations to preserve transparency while approaching the accuracy of strong tree-based predictors. A comprehensive empirical study on NYC taxi OD data shows PPML gravity as the most reliable physical baseline at hourly resolution, with AMBIT using POI-anchored or time-segment residuals delivering competitive performance and robust generalization under spatial holdouts. The work provides a reproducible pipeline, rich diagnostics, and actionable insights for urban decision-making, offering a practical path toward interpretable, scalable mobility modeling without sacrificing accuracy.

Abstract

Origin-destination (OD) flow prediction remains a core task in GIS and urban analytics, yet practical deployments face two conflicting needs: high accuracy and clear interpretability. This paper develops AMBIT, a gray-box framework that augments physical mobility baselines with interpretable tree models. We begin with a comprehensive audit of classical spatial interaction models on a year-long, hourly NYC taxi OD dataset. The audit shows that most physical models are fragile at this temporal resolution; PPML gravity is the strongest physical baseline, while constrained variants improve when calibrated on full OD margins but remain notably weaker. We then build residual learners on top of physical baselines using gradient-boosted trees and SHAP analysis, demonstrating that (i) physics-grounded residuals approach the accuracy of a strong tree-based predictor while retaining interpretable structure, and (ii) POI-anchored residuals are consistently competitive and most robust under spatial generalization. We provide a reproducible pipeline, rich diagnostics, and spatial error analysis designed for urban decision-making.
Paper Structure (58 sections, 2 equations, 6 figures, 30 tables)

This paper contains 58 sections, 2 equations, 6 figures, 30 tables.

Table of Contents

  1. Introduction
  2. Related work
  3. Spatial interaction baselines.
  4. Hybrid and deep OD models.
  5. Physics-enhanced residual learning.
  6. OD-as-node and completion-based approaches.
  7. PPML gravity and fixed effects.
  8. Monotone and interpretable tree models.
  9. Scale and distribution.
  10. Preprocessing.
  11. Distance definition.
  12. Features
  13. Methods
  14. Physical baselines
  15. PPML specification and margins.
  16. ...and 43 more sections

Figures (6)

  • Figure 1: SHAP summary for AMBIT residual (Gravity_POI baseline).
  • Figure 2: Spatial error diagnostics for the AMBIT residual model.
  • Figure 3: Metric comparison across physical models and tree-based baselines (CPC = common part of commuters).
  • Figure 4: sMAPE distribution across OD pairs (test set; sMAPE in $[0,2]$).
  • Figure A.1: Example-level SHAP waterfall plots for the AMBIT residual model.
  • ...and 1 more figures