Driving style recognition and comparisons among driving tasks based on driver behavior in the online car-hailing industry

Highlights

• Proposing a novel framework to classify driving styles based on large-scale online car-hailing data.

• Dividing the driving tasks into three categories (defined as cruising, ride requests, and drop-off) according to the tasks of professional drivers of online-hailed vehicles.

• Analyzing driving style based on maneuver detection (defined as turning, acceleration, deceleration).

• Analyzing variations in driving style during turning, acceleration, and deceleration maneuvers among the three driving tasks.

• The proposed framework is evaluated by a real case in Nanjing, China.

Abstract

As a product of the shared economy, online car-hailing platforms can be used effectively to help maximize resources and alleviate traffic congestion. The driver’s behavior is characterized by his or her driving style and plays an important role in traffic safety. This paper proposes a novel framework to classify driving styles (defined as aggressive, normal, and cautious) based on online car-hailing data to investigate the distinct characteristics of drivers when performing various driving tasks (defined as cruising, ride requests, and drop-off) and undergoing certain maneuvers (defined as turning, acceleration, and deceleration). The proposed model is constructed based on the detection and classification of driving maneuvers using a threshold-based endpoint detection approach, principal component analysis, and k-means clustering. The driving styles that the driver exhibits for the different driving tasks are compared and analyzed based on the classified maneuvers. The empirical results for Nanjing, China demonstrate that the proposed framework can detect driving maneuvers and classify driving styles accurately. Moreover, according to this framework, driving tasks lead to variations in driving style, and the variations in driving style during the different driving tasks differ significantly for turning, acceleration, and deceleration maneuvers.

Introduction

With the introduction of the ‘mobility as a service’ (MaaS) concept, internet companies have seized the opportunity to develop online car-hailing applications that provide quick travel options for the public. Characterized by environmentally-friendly and energy-saving features, online car-hailing already functions as an efficient tool to improve the utilization rate of private cars, create jobs, and alleviate traffic congestion in urban areas. However, the drivers of online-hailed cars (referred to simply as ‘drivers’ hereafter) have low bars for entry into online car-hailing jobs (Zhou et al., 2019). Also, road anger is common among these drivers (Feng et al., 2016). The car-hailing platform itself, which directs drivers who are waiting for ride requests, picking up passengers, and taking passengers to their destination, also influences the drivers’ driving styles, or driving habits, to some extent. To encourage these drivers to provide a safe and comfortable environment for passengers, the government and online car-hailing companies have invested significant resources into driver certification and online supervision for which recognition and analysis of the driver's driving style are important judgment factors.

The driving style of the driver is the main influential factor that determines a successful trip. Driving style plays an important role in vehicle energy management and driving safety, which in turn characterize safe (or unsafe) travel and the driving experience of passengers. Studies have shown that detecting driving style and providing drivers with feedback can help avoid unsafe driving behavior, reduce the frequency of traffic accidents, and improve traffic in general (Astarita et al., 2016; Hauber, 1980; Hickman and Geller, 2003; Taubman-Ben-Ari and Yehiel, 2012). In addition to affecting safety, driving style can significantly affect the fuel consumption of the vehicle (Meseguer et al., 2017). Moreover, because driving style affects vehicle maintenance costs, insurance costs, safety, etc., more and more research is focused on links between driving style and insurance premiums (Ellison et al., 2015; Kanarachos et al., 2018; Troncoso et al., 2011). In addition, driving style is the key to the development of driver assistance systems to improve the level of vehicle automation (Bellem et al., 2016; Karginova et al., 2012; Marina Martinez et al., 2018; Meiring and Myburgh, 2015; Sagberg et al., 2015). The safety and comfort of passengers depend on the driver’s driving style (Eboli et al., 2017). These core findings related to driving style have guided many scholars to research driving style recognition and classification.

The literature on driving style can be viewed from three aspects: definition, classification, and research methods. First, driving style can be defined in multiple ways. For example, Ishibashi et al. (2007) considered driving style to be the driver's attitude towards thinking about driving tasks. Elander et al. (1993) believed that driving style is each driver’s thinking about ways to drive. Dörr et al. (2014) proposed a practical description that defines driving style as the way to complete a driving task. Johnson and Trivedi (2011) believed that acceleration, deceleration, turning, and lane changing can be used to identify driving styles.

Previous studies also have classified driving styles in different ways. For simulated scenarios, Bär et al. (2011) classified driving styles into five categories: aggressive, anxious, economical, keen, and sedate. Aljaafreh et al. (2012) used fuzzy logic models to classify driving styles into four categories based on acceleration data: below normal, normal, aggressive, and very aggressive. Johnson and Trivedi (2011) used smartphones as sensors to classify driving characteristics into two categories: non-aggressive and aggressive. However, most studies classify driving styles into three categories: aggressive, normal, and cautious (Dörr et al., 2014; Higgs and Abbas, 2013; Xu et al., 2015). We selected these three categories to classify driving styles in this study.

In recent years, scholars worldwide have made substantial efforts to study driving styles in various ways. For example, Ishibashi et al. (2007) proposed a driving style questionnaire to extract key factors from self-reports and describe different styles. Other researchers have used the Multidimensional Driving Style Inventory to assess four broad domains of driving style (Taubman-Ben-Ari et al., 2004; Taubman-Ben-Ari and Skvirsky, 2016). However, responses to this questionnaire often reflected the subjective judgment of the participants which may deviate from their actual performance on the road. Some researchers have analyzed driving style by conducting simulation experiments (Bär et al., 2011; Chen et al., 2013; Doshi and Trivedi, 2010). Such simulations have shown that drivers can be led to exhibit different driving styles by manually controlling the driving environment. However, the problem with this simulation method is that no clear distinctions are evident among various driving styles.

Other researchers have collected realistic driving data from equipment installed on vehicles for analysis purposes (Boyce and Geller, 2002; Kleisen, 2013; Ma et al., 2020, 2019). For example, Constantinescu (2010) used hierarchical cluster analysis and principal component analysis (PCA) to analyze GPS tracking data, such as vehicle speed and acceleration, to classify driving styles into five categories. Van Ly et al. (2013a) used a clustering algorithm and support vector machine model to develop a driving style classification system based on the data collected during multiple trips of two drivers using a vehicle with an inertial sensor. Qi et al. (2019) proposed two topic models (mLDA and mHLDA) to obtain distinguishable driving style information with hidden structures from real-world driving scenarios. Van Ly et al. (2013a) used data collected by a smartphone and its inertial detector to study whether a driver's driving maneuvers could uniquely identify the driver.

Research into driving styles of drivers of online-hailed cars that is based on driving tasks is limited. Most previous studies have tended to assume that the driver's driving style is fixed within each trip, which ignores the fact that driving style may change during a single trip. For example, Feng et al. (2018) used support vector clustering based on detected maneuvers to classify driving style and found that the driving style of a single driver is not consistent and may vary within a single trip. The traffic environment and external conditions also can affect the driver’s behavior; that is, even drivers who typically exhibit a normal and safe driving style may change to an aggressive driving style when they are required to complete a task within a limited time or encounter congested traffic or interference from other drivers. Therefore, dividing a journey into several driving tasks based on changes in the driver's physiological state may be a more reasonable way to investigate driving style.

This paper puts forward a new framework to analyze changes in driving style that take place during different driving tasks for online car-hailing platforms, which can identify driving maneuvers and classify driving styles. This paper is structured as follows. In Section 2, the data sources and tests are discussed in detail. In Section 3, the four-part framework for driving style classifications based on online car hailing data is proposed. The results for each part are presented and discussed in Section 4. Next, the variability in driving style during different driving tasks is analyzed based on the classified maneuvers. Finally, the main findings are summarized and future work is discussed in Section 5.