The Impact of driver distraction and secondary tasks with and without other co-occurring driving behaviors on the level of road traffic crashes

Abstract

Driving safety is typically affected by concurrent non-driving tasks. These activities might negatively impact the trips’ outcome and cause near-crash or crash incidents and accidents. The crashes impose a tremendous social and economic cost to society and might affect the involving individuals’ quality of life. As it stands, road injuries are ranked among top-ten leading causes of death by the World Health Organization. Distracted driving is defined as an attention diversion of the driver toward a competing activity. It was shown in numerous studies that distracted driving increase the probability of near-crash or crash events. By leveraging the statistical power of the large SHRP2 naturalistic data, we are able to quantify the preponderance of specific distractions during daily trips and confirm the causality factor of an ubiquitous non-driving task in the crash event. We show that, except for phone usage which happens more frequently in near-crash and crash categories than in baseline trips, both distracted driving and secondary tasks occur almost uniformly in different types of trips. In this study, we investigate the impact of the co-occurrence of distracted driving with other driving behaviors and secondary tasks. It is found that the co-occurrence of distracted driving with other driving behaviors or secondary tasks increase the chance of near-crash and crash events. This study's findings can inform the design and development of more precise and reliable driving assistance and warning systems.

Introduction

According to NHTSA, 25 percent of the police-reported crashes are due to driver inattention defined as “insufficient or no attention to activities critical for safe driving” (Regan et al., 2011). The most substantial form of driver inattention is distracted driving. Distracted driving is defined as events or activities within or outside the vehicle (Young et al., 2007) that negatively affect a driver's ability to process information that is necessary to operate a vehicle safely (Regan and Hallett, 2011, Regan et al., 2008, Regan et al., 2011). This includes talking on cell phones, texting, eating, drinking, and other non-driving activities, called secondary tasks (Regan and Hallett, 2011). Distracted driving accounts for approximately 16 percent of economic loss and 15 percent of societal harm. In addition, 10 percent of fatal and 18 percent of injury crashes have been reported as “distraction-affected crashes” (Blincoe et al., 2015). These numbers represent driving trips that ended in non-fatal injuries or deaths. However, it is shown that as much as 16.1% of driving time gets affected by inattention (Stutts et al., 2003). Moreover, distracted driving has adverse impacts on traffic operation due to greater fluctuation in speed and significant lane deviation (Stavrinos et al., 2013). Therefore, numerous research studies have focused on the definition, theoretical foundation, formulation, prediction, and prevention of distraction and distracted driving to inform the development of the technological, behavioral, and infrastructure mitigating measures to enhance driving safety.

Classically, studies that focus on distracted driving use different combinations of data collection and analytical approaches. For example, the data used for examining distracted driving may be collected from human-in-the-loop simulation studies for retrospective (Jin et al., 2012, Ameyoe et al., 2015, Stavrinos et al., 2013) and real-time analysis or prediction (Wang et al., 2015, Liang et al., 2007). Another common approach is collecting naturalistic driving data using an instrumented vehicle for retrospective (Dukic et al., 2013, Jenkins et al., 2017, Aksan et al., 2013, Li et al., 2018) or real-time analysis and prediction (Liu et al., 2016, Deshmukh and Dehzangi, 2017, Kircher and Ahlstrom, 2010, Botta et al., 2019). Another approach is adopting qualitative techniques for data collection, such as interviews (Bakiri et al., 2013). Different methods are also used for analysis, modeling, and prediction of distraction, such as driver modeling including perceptual and motor components (Hermannstädter and Yang, 2013, Ameyoe et al., 2015, Li et al., 2018), statistical analysis (Bakiri et al., 2013, Dukic et al., 2013), and machine learning algorithms such as classification and regression (Jin et al., 2012, Liu et al., 2016, Jenkins et al., 2017, Deshmukh and Dehzangi, 2017, Wang et al., 2015, Kircher and Ahlstrom, 2010, Liang et al., 2007, Botta et al., 2019).

An avenue of research on distracted driving now highlights the effects of secondary tasks. According to the Second Strategic Highway Research Program (SHRP 2) Researcher Dictionary for Video Data Reduction (VTTI, 2015), the secondary task is defined as any distraction that includes non-driving related glances away from the direction of vehicle movements such as radio adjustments, seatbelt adjustments, window adjustments, visor adjustment, and other non-critical tasks. It does not include tasks that are critical to the driving, such as speedometer checks, blind spot checks, activating wipers/headlights, and other critical tasks.

As a specific data collection approach gets adopted, different sets of variables get generated. These variables can be grouped into three categories. The first category includes variables related to the driver (such as age and prior experience, and visual, motor, and cognitive capabilities) or variables measured to collect the level of distraction or inattention of driver (such as physiological changes in the driver state, eye movement patterns, and brain activity measures). The second category includes variables collected from the instrumented vehicle or simulator dynamics (such as lateral and longitudinal speed and acceleration, lateral deviation, and steering angle over the course of driving). Then the third category is composed of variables associated with the environment. The latter category characterizes the sources of internal or external distractions such as cell-phone and billboards or time and physical characteristics of the environment, such as traffic signs, the surrounding vehicle dynamics, or road curvature.

The findings of the above studies can be summarized as follows. In the studies conducted to measure the impacts of distracted driving, it is shown that distracted driving significantly and adversely impact the performance of drivers. Besides, distracted drivers experience changes in their physiological and brain state, functionality, and performance. These changes are meaningful enough to be used for prediction purposes and for the design and development of warning systems.

However, the impact of distraction co-occurring with other driving behaviors are studied only in a few studies. One study shows the relationship between driving drowsy and distracted driving (Anderson and Horne, 2013), while another study investigates the distractive effects of cell phone use on safe driving (Unknown, 2003). None of the previously reported studies explicitly considered the data-driven co-occurrence of driving behaviors and secondary tasks to crash risk. In this study, we categorize driving epochs based on their outcome: (i) epochs ending in a crash, (ii) epochs with a near-crash incident (but no crash), and (iii) baseline epochs without any near-crash or crash incidents. We adopt a data-driven approach to identify co-occurring behaviors in a repository of driving behaviors. The data used in this study is collected in a naturalistic driving experiment. The objectives of this study are to identify: frequent driving behaviors and their co-occurrences, frequent secondary tasks and their co-occurrences with driving behaviors, and impacts and frequency of distraction or secondary tasks with and without other driving behaviors among different outcome categories.

To meet the objectives of this study, we mine frequent driving behaviors and secondary tasks in a data set. The association rule mining technique is utilized to identify frequent driving behaviors. This approach has been adopted for mining of co-occurring patterns in other applications in previous studies as well. For example, in Brossette et al. (1998), a data set of health surveillance data is mined to reveal unknown patterns. Or, Abdullah et al. (2008) uses a data set of medical billing data to identify frequent associations between diagnosis codes and treatment procedures. A similar approach is adopted in Shan et al. (2008), Kareem et al. (2017) for the identification of suspicious claims and potentially fraudulent individuals from billing records. Other applications are prediction and forecasting of cardiovascular diseases and heart attacks (Ordonez, 2006, Jabbar et al., 2011, Khare and Gupta, 2016), location-wise and time-wise mining of frequent diseases (Ilayaraja and Meyyappan, 2013), identification of associations among environmental exposure to different chemical compounds and adverse health outcomes (Bell and Edwards, 2014), and identification of patterns which can inform the diagnosis of asthma in pediatric using a sequential version of this method (Campbell et al., 2020). For a survey on the applications of association rule mining techniques in healthcare applications, please refer to Altaf et al. (2017).

Using the association rule mining technique, this study aims to identify frequent co-occurring behaviors. In other words, we would like to investigate and identify the non-driving behaviors, more specifically, distraction, and related secondary tasks, that are commonly observed in different types of epochs. At the same time, their co-occurrences are more frequent in near-crash or crash epochs. Overlooking potential differences in impacts of behaviors when they occur individually or co-occur with other behaviors might result in higher false positive and false negative rates. Therefore, identifying such sets of behaviors can inform more accurate predictions of epochs’ outcomes and, consequently, the design and development of more reliable warning systems.