Temporal analysis of driving efficiency using smartphone data

https://doi.org/10.1016/j.aap.2021.106081Get rights and content

Highlights

  • The temporal evolution of driving safety efficiency is investigated.

  • Safety efficiency is evaluated based on driving data collected from smartphones.

  • Efficiency is studied considering speeding, distraction, and aggressiveness in all road types.

  • Three main driving groups are identified namely, moderate drivers, unstable drivers and cautious drivers.

  • The characteristics of each driving group of drivers are presented and discussed.

Abstract

This paper attempts to shed light on the temporal evolution of driving safety efficiency with the aim to acquire insights useful for both driving behavior and road safety improvement. Data exploited herein are collected from a sophisticated platform that uses smartphone device sensors during a naturalistic driving experiment, at which the driving behavior from a sample of two hundred (200) drivers during 7-months is continuously recorded in real time. The main driving behavior analytics taken into consideration for the driving assessment include distance travelled, acceleration, braking, speed and smartphone usage. The analysis is performed using statistical, optimization and machine learning techniques. The driver’s safety efficiency index is estimated both in total and in several consecutive time windows to allow for the investigation of safety efficiency evolution in time. Initial data analysis results to the most critical components of microscopic driving behaviour evolution, which are used as inputs in the k-means algorithm to perform the clustering analysis. The main driving characteristics of each cluster are identified and lead to the conclusion that there are three main driving groups of the a) moderate drivers, b) unstable drivers and c) cautious drivers.

Introduction

Many studies in driving behavior literature (Tselentis et al., 2019; Matthews et al., 1996, 1998; Young et al., 2011) have focused on measuring driving safety efficiency. Driving safety efficiency in this study refers to the amount of driving events (harsh braking, harsh acceleration, mobile phone usage, speeding) occurred during a certain driving period (Tselentis et al., 2019). The most efficient drivers are those with the least number of events. Nonetheless, only a few of them have demonstrated that there is a potential in analyzing and evaluating driving behaviour using microscopic driving data (e.g. driving over the speed limits, mobile phone usage and the number of harsh acceleration and braking events occurred while driving) collected from naturalistic driving experiments (Tselentis et al., 2019), (Toledo et al., 2008; Eboli et al., 2016, 2017; Guo and Fang, 2013; Dingus et al., 2016; Mantouka et al., 2020). It is extremely important from a safety perspective to identify the behavioral parameters that influence drivers and, therefore, the probability of getting involved in a crash.

Several studies, thus far, have investigated the microscopic driving factors that could potentially be incorporated in driving assessment models as well as the methodologies for driving behavior data collection and analysis (Tselentis et al., 2019, 2017). Driving behavior is complex; it can be influenced by the mode of transport (Barmpounakis and Vlahogianni, 2020), and the type of road (Stavrakaki et al., 2020). Due to this inherent complexity, a different optimal driving policy may exist for each driver (Vlachogiannis et al., 2020). Finally, driving behavior, if controlled, can lead not only to safer roads, but also to healthier cities (Adamidis et al., 2020).

As for the data collection, literature review revealed that the methodologies most commonly used include driving simulators (Desmond et al., 1998; Lenné et al., 1997), questionnaires (Matthews et al., 1998) combined with simulators and naturalistic driving experiments (Toledo et al., 2008; Birrell et al., 2014; Hanowski et al., 2003). Naturalistic experiments provide a wide perspective of understanding typical microscopic travel and driving behaviour (Mantouka et al., 2020). A naturalistic study can help (Regan et al., 2012): a) determine crash risk, b) study the interaction between road/ traffic conditions and driver’s behaviour, c) understand the interaction between car drivers and vulnerable road users, d) specify the relationship between driving pattern and vehicle emissions and fuel consumption, and many other aspects of traffic participation. The most popular devices for monitoring driving measures are recorders related to the car engine (Zaldivar and Calafate, 2011; Backer-Grøndahl and Sagberg, 2011) such as on-board-diagnostics (OBD) devices and smartphones (Vlahogianni and Barmpounakis, 2017).

Driving efficiency assessment using microscopic driving parameters is thoroughly studied (Tselentis et al., 2019). Nevertheless, the evolution of the driving performance in time (in the long term) has not yet been investigated. The temporal characteristics of driving efficiency and especially stationarity, trend and volatility, are of outmost importance when driving efficiency is measured. This is because the average driving efficiency might be representative of the total driving risk only in those cases when driving behavior is not fluctuating in time. This occurs solely when a driver retains a steady behavior that is usually close to the average and is not significantly changing while the driver is being monitored. Therefore, driving efficiency measured might not be representative of a driver’s risk in a case of an unstable driver whose behaviour is volatile. In other words, the methodology of average efficiency estimation might not always be able to stand alone. For instance, when comparing the behavior of two drivers of the same average efficiency, it is more likely for the less steady driver (the one with the more volatile behavior) to have a higher crash risk since he features a higher number of less efficient trips.

In order to fill this research gap, this paper presents a methodological framework to study the temporal evolution of measured driver’s efficiency with the aim to provide insights on the different driving behavior profiles. Whereas the authors’ previous work (Tselentis et al., 2019) focused on assessing the aggregated driving behavior, this is an effort to reveal more information on how this behavior evolves over time, providing thus more information on the dynamic aspects of driving behavior. In other words, instead of providing only one driving efficiency index for the total monitoring period, monitoring period is divided into several shorter consecutive time periods and an individual driving efficiency index is provided for each of these periods. This process creates a time-series of driving efficiency indices for each driver, the trend and volatility of which is estimated. The latter two characteristics of the time-series, together with the driving efficiency of the total monitoring period, are used as inputs in a clustering algorithm to identify prevailing driver profiles. These profiles constitute an attempt to recognize the several existing driver profiles and their characteristics. The main driving characteristics of each group are presented and important conclusions are drawn regarding the features of each driving group. This methodology could be exploited as a platform’s service in order to provide recommendations to drivers on how to improve their driving efficiency and become less risky.

Section snippets

Methods for Driving Efficiency Analysis

Previous research on efficiency analysis has shown that data envelopment analysis (DEA) is an effective methodology to measure driving efficiency (Shone, 1981; Emrouznejad et al., 2008). Despite the fact that DEA is mostly used in business, economics, management and health (Cook and Seiford, 2009; Hollingsworth et al., 1999), it has also been implemented in transport fields in assessing public transportation system performance (Karlaftis et al., 2013), as well as traffic safety studies (Egilmez

Driving safety efficiency estimation - mathematical formulation

For the sake of simplicity, it is noted that from now on DMUs will be referred as drivers. In order to evaluate the driving efficiency of Driver0 and assuming a sample of N drivers, let X and Y represent the set of inputs and outputs respectively, for the rest of the drivers’ sample. In other words, X={x1,x2,...,xi} and Y={y1,y2,...,yi} where i1,N1. The input-oriented CCR model evaluates the efficiency of Driver0 by solving the linear program (Ramanathan, 2003) presented below. Considering

Experimental Data Collection

OSeven Telematics (OSeven, 2021) has developed an integrated platform for the recording, transmission, storage, evaluation and visualization of driving behaviour data using a smartphone application, statistical and advanced machine learning (ML) algorithms. Recorded data come from various smartphone sensors (including the accelerometer, the gyroscope, the GPS and the compass) and data fusion algorithms provided by Android and iOS. The application transmits all data to a central database after

Components of the efficiency time series

Driver’s efficiency is estimated for each time step of a sliding time window, following exactly the same procedure described above in the estimation of total driving efficiency. This allows for studying the evolution of the average driving efficiency over different timeframes from the beginning of the recording time until the end of each timeframe. The length of the time window is estimated using specific statistical tests that identify the convergence of the driving analytics of each driver to

Discussion

This paper provides a structured approach to investigate the evolution of driving efficiency in time, aiming to draw conclusions on the different existing driving patterns. Driving efficiency estimation is based on a methodology introduced in previous research using DEA ((Tselentis et al., 2019)). Based on driving analytics of a sample of two hundred (200) drivers during 7-months, the analysis of the efficiency time series arising revealed that although there is a higher range of volatility in

CRediT authorship contribution statement

Dimitrios I. Tselentis: Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Project administration, Software, Validation, Visualization, Writing - original draft, Writing - review & editing. Eleni I. Vlahogianni: Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Project administration, Supervision, Validation, Writing - review & editing. George Yannis: Conceptualization, Investigation, Methodology, Project administration, Supervision,

Declaration of Competing Interest

We would like to inform you that there are no conflicts of interests to be declared. We remain at your disposal for any further information.

Acknowledgement

The authors would like to thank OSeven Telematics, for providing all necessary data to accomplish this study.

References (50)

  • F. Guo et al.

    Individual driver risk assessment using naturalistic driving data

    Accid. Anal. Prev.

    (2013)
  • R.J. Hanowski et al.

    An on-road study to investigate fatigue in local/short haul trucking

    Accid. Anal. Prev.

    (2003)
  • Michael G. Lenné et al.

    Time of day variations in driving performance

    Accid. Anal. Prev.

    (1997)
  • E.G. Mantouka et al.

    Identifying driving safety profiles from smartphone data using unsupervised learning

    Saf. Sci.

    (2019)
  • R. Shichrur et al.

    Determining the sampling time frame for in-vehicle data recorder measurement in assessing drivers

    Transp. Res. Part C Emerg. Technol.

    (2014)
  • Tomer Toledo et al.

    In-vehicle data recorders for monitoring and feedback on drivers’ behavior

    Transp. Res. Part C Emerg. Technol.

    (2008)
  • Dimitrios I. Tselentis et al.

    Innovative motor insurance schemes: a review of current practices and emerging challenges

    Accid. Anal. Prev.

    (2017)
  • D.I. Tselentis et al.

    Driving safety efficiency benchmarking using smartphone data

    Transp. Res. Part C Emerg. Technol.

    (2019)
  • Eleni I. Vlahogianni et al.

    Driving analytics using smartphones: algorithms, comparisons and challenges

    Transp. Res. Part C Emerg. Technol.

    (2017)
  • Mark S. Young et al.

    Safe driving in a green world: a review of driver performance benchmarks and technologies to support ‘smart’driving

    Appl. Ergon.

    (2011)
  • F.K. Adamidis et al.

    Effects of controlling aggressive driving behavior on network-wide traffic flow and emissions

    Int. J. Transp. Sci. Technol.

    (2020)
  • E. Barmpounakis et al.

    Powered Two-Wheeler detection using crowdsourced smartphone data

    Ieee Trans. Intell. Veh.

    (2020)
  • Stewart A. Birrell et al.

    Effect of using an in-vehicle smart driving aid on real-world driver performance

    Ieee Trans. Intell. Transp. Syst.

    (2014)
  • S. Cameron

    Making regression analysis more useful, II"

    Econometrics. Maidenhead: McGraw Hill Higher Education.

    (2005)
  • W.W. Cooper et al.

    Data envelopment analysis: a comprehensive text with models, applications, references and DEA-Solver Software

    Springer Science & Business Media

    (2006)
  • Cited by (14)

    • Analysis of mobile phone use engagement during naturalistic driving through explainable imbalanced machine learning

      2023, Accident Analysis and Prevention
      Citation Excerpt :

      The high penetration of smartphones worldwide (GSM Association, 2021) as well as the low cost and ease of deployment allow for new opportunities in terms of data collection, storage, management and procession (Vlahogianni and Barmpounakis, 2017; Chan et al., 2019). A number of recent studies have used data from smartphone sensors for driving behavior monitoring and analysis in terms of investigating risk factors (Paefgen et al., 2012; Ma et al., 2017; Petraki et al., 2020), identifying driving styles through clustering techniques (Meseguer et al., 2013; Mantouka et al., 2019; Tselentis et al., 2021) or even evaluating driver feedback (Toledo and Shiftan, 2016; Dahlinger et al., 2018; Kontaxi et al., 2022). Mobile phone has emerged as a reliable in-vehicle device for collecting high-resolution naturalistic driving data (Singh & Kathuria, 2021) and smartphone behavior monitoring are part of a telematics-based revolution which has transformed the insurance industry (Ziakopoulos et al., 2022b).

    • Assessing safety critical driving patterns of heavy passenger vehicle drivers using instrumented vehicle data – An unsupervised approach

      2021, Accident Analysis and Prevention
      Citation Excerpt :

      Similarly, Tselentis et al. (2019) classified drivers into three categories: non-efficient, weakly efficient, and most efficient based on the abstract driving performance exhibited over the study period. The authors further investigated the driving safety efficiency evolution in time and categorized drivers based on the macroscopic temporal driving characteristics (Tselentis et al., 2021). Yao et al. (2019) clustered the left turn data corresponding to six similar curves using DTW and HMM (Hidden Markov Model).

    View all citing articles on Scopus
    View full text