The Analysis of Classification and Spatiotemporal Distribution Characteristics of Ride-Hailing Driver’s Driving Style: A Case Study in China

Runkun Liu, Haiyang Yu, Yilong Ren, Shuai LiuInternational Journal of Environmental Research and Public Health 2022 · DOI

TL;DR

This study analyzes the driving styles of over 34,000 ride-hailing drivers in China, revealing that most drivers exhibit varying driving behaviors throughout the day and across different road environments (segments vs. intersections). The findings highlight the need for dynamic, context-aware driver management and training strategies to improve road safety.

What Problem Does This Paper Address?

Ride-hailing services have become ubiquitous, but the driving styles of their drivers can vary significantly, posing potential risks to safety and service quality. While previous research has attempted to classify driving styles, it often overlooked the crucial temporal (time of day) and spatial (road segments vs. intersections) variations in these behaviors. This lack of understanding hinders the development of effective, targeted training programs and management policies for ride-hailing platforms and authorities. The paper aims to fill this gap by analyzing how driving styles change over time and in different driving environments.

Methods

The researchers employed a data-driven approach using connected vehicle trajectory data.

  • Feature Extraction: Six key features were extracted to represent driving style: mean and standard deviation of speed ($\mu_s, \sigma_s$), acceleration ($\mu_a, \sigma_a$), and deceleration ($\mu_d, \sigma_d$).
  • Clustering: The k-means clustering algorithm was used to group drivers into distinct driving style categories. The elbow method was employed to determine the optimal number of clusters, which was found to be three.
  • Temporal Analysis: The day was divided into three periods (rush hour, flat period, night period) based on speed distributions to analyze how driving styles vary across these times.
  • Spatial Analysis: The road network was segmented into "road segments" and "intersections" to investigate how driving styles differ between these two environments. Lognormal and exponential decay models were used to fit speed and acceleration/deceleration distributions, respectively.

Datasets

The study utilized trajectory data from the Didi Chuxing Global Astrometric Interferometer for Astrophysics (GAIA) Open Dataset Initiative. This dataset comprises approximately 31 million records from over 34,000 ride-hailing drivers in Chengdu, China, recorded every 3 seconds on November 1, 2016. Intersection location data was also obtained from Baidu Location-Based Services.

Key Insights and Ideas

  • Dynamic Driving Styles: The assumption that a driver's style is static is flawed. Most ride-hailing drivers adapt their behavior based on the time of day and the driving environment.
  • Intersections as High-Risk Zones: Intersections consistently show higher acceleration and deceleration values compared to road segments across all driver types, suggesting they are inherently more dangerous environments.
  • Aggressive Drivers on Open Roads, Cautious at Intersections: Aggressive drivers tend to exhibit their riskier behaviors (higher speeds) on road segments, while conservative drivers adopt more cautious strategies at intersections.
  • The "Unstable" Driver: A small but significant portion of drivers exhibit substantial shifts in their driving style, indicating potential instability that warrants attention.
  • Policy Implications: The findings strongly support the need for dynamic risk assessment and tailored interventions for ride-hailing drivers, moving beyond a one-size-fits-all approach.

Key Results

  • Driving Style Classification: Drivers were classified into three types:
  • Type A (Aggressive): Highest speed, acceleration, and deceleration parameters (19% of drivers).
  • Type B (Normal): Moderate parameters (54% of drivers).
  • Type C (Cautious): Lowest parameters (27% of drivers).
  • Temporal Variation: Only 31.79% of drivers maintained a stable driving style throughout the day. The proportion of aggressive drivers was notably higher during the night period.
  • Spatial Variation:
    • Drivers generally maintained higher speeds on road segments compared to intersections.
  • Acceleration and deceleration were consistently higher at intersections than on road segments for all driver types.
  • Aggressive drivers showed more aggressive driving styles on road segments, while conservative drivers were more conservative at intersections.

Limitations and Future Work

  • Single Day Data: The study relied on data from a single day, which may not fully capture long-term driving behavior patterns. Future work could incorporate data from weeks, months, or even years.
  • Limited Driver Information: The analysis did not incorporate individual driver characteristics beyond trajectory data. Future research could explore the influence of factors like driving experience, personality, and psychological states.
  • Geographical and Population Factors: The study focused on Chengdu and did not extensively consider geographical factors or the impact of floating populations, which could influence driving behavior.
  • Causality of Style Changes: While the study identified style changes and their correlation with traffic conditions, it did not delve into the underlying physiological or psychological reasons for these shifts. Future work could investigate these aspects through experiments and surveys.

Summary updated

Abstract

Monitoring the driving styles of ride-hailing drivers is helpful for providing targeted training for drivers and improving the safety of the service. However, previous studies have lacked analyses of the temporal variation as well as spatial variation characteristics of driving styles. Understanding the variations can also help authorities formulate driver management policies. In this study, trajectory data are used to analyze driving styles in various temporal and spatial scenarios involving 34,167 drivers. The k-means method is used to cluster sample drivers. In terms of driving style time-varying, we found that only 31.79% of drivers could maintain a stable driving style throughout the day. Spatially, we divided the research area into two parts, namely, road segments and intersections, to analyze the spatial driving characteristics of drivers with different styles. The speed distribution, the acceleration and deceleration distributions are analyzed, results indicated that aggressive drivers display more aggressive driving styles in road segments, and conservative drivers exhibit more conservative driving styles at intersections. The findings of this study provide an understanding of temporal and spatial driving behavior factors for ride-hailing drivers and offer valuable contributions to ride-hailing driver training and road safety management.
BibTeX
@article{liu2022analysis,
  author = {Runkun Liu and Haiyang Yu and Yilong Ren and Shuai Liu},
  title = {{The Analysis of Classification and Spatiotemporal Distribution Characteristics of Ride-Hailing Driver’s Driving Style: A Case Study in China}},
  journal = {International Journal of Environmental Research and Public Health},
  year = {2022},
  doi = {10.3390/ijerph19159734},
}
Figure