Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Frailty-Aware Transformer for Recurrent Survival Modeling of Driver Retention in Ride-Hailing Platforms

Shuoyan Xu, Yu Zhang, Eric J. Miller

TL;DR

The paper addresses driver retention in ride-hailing by formulating idle-period termination as a recurrent survival problem. It introduces Frailty-Aware Cox Transformer (FACT), which combines a Transformer encoder with causal masking and driver-specific frailty embeddings to predict time-to-idle termination across repeated events. Empirical results on Toronto data show FACT achieving top time-dependent discrimination (C-index up to ~0.721) and calibration (low IBS), outperforming classical and deep-learning baselines. The approach enables personalized risk estimation and actionable retention strategies, with potential extensions to real-time signals and policy evaluations.

Abstract

Ride-hailing platforms are characterized by high-frequency, behavior-driven environments. Although survival analysis has been applied to recurrent events in other domains, its use in modeling ride-hailing driver behavior remains largely unexplored. This study formulates idle behavior as a recurrent survival process using large-scale platform data and proposes a Transformer-based framework that captures long-term temporal dependencies with causal masking and incorporates driver-specific embeddings to model latent heterogeneity. Results on Toronto ride-hailing data demonstrate that the proposed Frailty-Aware Cox Transformer (FACT) achieves the highest time-dependent C-indices and lowest Brier Scores, outperforming classical and deep learning survival models. This approach enables more accurate risk estimation, supports platform retention strategies, and provides policy-relevant insights.

Frailty-Aware Transformer for Recurrent Survival Modeling of Driver Retention in Ride-Hailing Platforms

TL;DR

The paper addresses driver retention in ride-hailing by formulating idle-period termination as a recurrent survival problem. It introduces Frailty-Aware Cox Transformer (FACT), which combines a Transformer encoder with causal masking and driver-specific frailty embeddings to predict time-to-idle termination across repeated events. Empirical results on Toronto data show FACT achieving top time-dependent discrimination (C-index up to ~0.721) and calibration (low IBS), outperforming classical and deep-learning baselines. The approach enables personalized risk estimation and actionable retention strategies, with potential extensions to real-time signals and policy evaluations.

Abstract

Ride-hailing platforms are characterized by high-frequency, behavior-driven environments. Although survival analysis has been applied to recurrent events in other domains, its use in modeling ride-hailing driver behavior remains largely unexplored. This study formulates idle behavior as a recurrent survival process using large-scale platform data and proposes a Transformer-based framework that captures long-term temporal dependencies with causal masking and incorporates driver-specific embeddings to model latent heterogeneity. Results on Toronto ride-hailing data demonstrate that the proposed Frailty-Aware Cox Transformer (FACT) achieves the highest time-dependent C-indices and lowest Brier Scores, outperforming classical and deep learning survival models. This approach enables more accurate risk estimation, supports platform retention strategies, and provides policy-relevant insights.

Paper Structure

This paper contains 30 sections, 18 equations, 8 figures, 6 tables.

Table of Contents

  1. Introduction
  2. Literature Review
  3. Driver behavior and retention on ride-hailing platforms
  4. Survival analysis, recurrent event and ride-hailing
  5. Deep Learning for Survival Analysis
  6. Methodology
  7. Problem Definition
  8. Frailty-aware Transformer-Based Risk Function
  9. Loss Function for Recurrent Events
  10. Evaluation Metrics
  11. Concordance Index (C-index)
  12. Brier Score
  13. Data
  14. Data Source
  15. Ride-Hailing Records
  16. ...and 15 more sections

Figures (8)

  • Figure 1: Illustration of censored and complete events defined in ride-hailing
  • Figure 2: Framework of the proposed model
  • Figure 3: Study Area
  • Figure 4: Spatiotemporal distribution of key supply-side indicators in Toronto’s ride-hailing system.
  • Figure 5: Illustration of censored and complete events defined in ride-hailing
  • ...and 3 more figures