Analyzing the Impact of C-ITS Services on Driving Behavior: A Case Study of the Daejeon–Sejong C-ITS Pilot Project in South Korea

Abstract

This paper analyzes the impact of C-ITS service on driving behavior, focusing on a pilot project in Daejeon–Sejong, South Korea. C-ITS, an advanced technology, enables bidirectional wireless communication between vehicles or infrastructure, allowing for real-time traffic data collection and dissemination. The study uses a unique analytical method, employing parallel processing techniques for variable extraction and a paired t-test to examine the short-term effects of C-ITS on driving behavior. Findings indicate a significant change in drivers’ behavior, particularly in average speed, hard braking rate, severe deceleration rate, speeding rate, and excessive speeding rate, towards safer trends after receiving C-ITS warning services. Reductions in hard braking and severe deceleration were immediate after C-ITS service initiation, while a decrease in excessive speeding was observed after four months. Further research is needed to identify other potential influencing variables and provide an unbiased evaluation of C-ITS effectiveness. The study’s implications highlight its role in promoting public acceptance of C-ITS-service-based cooperative autonomous driving strategies.

1. Introduction

C-ITS (Cooperative Intelligent Transportation System) is centered around an advanced communication method that goes beyond conventional ITS (Intelligent Transportation System), which typically relies on one-way traffic information collection. C-ITS utilizes two-way wireless communication, specifically focusing on the WAVE (Wireless Access in Vehicular Environments) communication protocol, to enable interactions between vehicles (V2V) and between vehicles and infrastructure (V2I). WAVE is a specialized set of standards that facilitates the seamless exchange of information in vehicular networks, forming the core of the broader concept known as Vehicle-to-Everything (V2X). This includes communication with traffic signals, roadside units, and other road users. WAVE operates in the 5.9 GHz band and is designed to support short-range, high-speed wireless communication, making it particularly suitable for dynamic and rapidly changing traffic environments. The implementation of WAVE in C-ITS plays a vital role in providing real-time traffic situation information, enhancing road safety, and paving the way for future automated driving systems. By leveraging the WAVE communication method, C-ITS is able to offer more responsive and adaptive solutions to traffic management and control, reflecting the ongoing evolution of intelligent transportation technologies (Figure 1).

In conventional ITS, the collection and provision of traffic information relies on specific locations where traffic information collection and dissemination devices were installed. As a result, it is challenging to respond promptly to unexpected traffic situations. In contrast, C-ITS enables the real-time collection of various traffic information remotely through terminals and provides that information to other drivers, facilitating rapid and safe responses to sudden incidents [1,2]. By identifying potential hazards and preventing accidents in advance, C-ITS distinguishes itself from traditional ITS and is therefore an essential technology for creating a future autonomous driving environment [3,4,5,6].

C-ITS has garnered significant attention in the field of smart city as a cutting-edge technology that contributes to the reduction in traffic accidents, alleviation of congestion, and achievement of carbon reduction goals [7,8,9,10]. Among regions, Europe has emerged as the front-runner in the development of C-ITS technology. Since the mid-2000s, under the guidance of the European Union, numerous pilot projects have been conducted to validate C-ITS technology [11,12,13,14,15,16,17]. Since the mid-2000s, under the guidance of the European Union, numerous pilot projects have been conducted to validate C-ITS technology. Representative examples of such C-ITS deployment projects include C-Mobiles and C-Road [18,19,20,21]. These projects aim to achieve interoperable deployment by listing services based on their technical maturity and social benefit.

The services expected to be available in the short term are referred to as Day 1 and Day 1.5 services [20]. Lu et al. (2018) also provided a list of Day 1 and Day 1.5 services, among which the Green Light Optimal Speed Advisory (GLOSA) stands out as one of the most well-known Day 1 services [22]. GLOSA, as a Day 1 service, offers drivers, whether they are human or autonomous, guidance on maintaining an appropriate speed to reach upcoming intersections while the traffic signals remain green. This information is communicated through interactions between the vehicle and the traffic lights, utilizing the on-board unit (OBU) in the vehicle [20]. By utilizing the GLOSA service, drivers can optimize their speed and avoid unnecessary stops at intersections, thus enhancing traffic efficiency and reducing delays [23].

The majority of C-ITS-related journal papers published since 2020 are founded upon empirical projects carried out in European countries [3,24,25,26]. Kotsi et al. (2020) offer an overview of C-ITS milestones in both Europe and the United States [27]. Currently, Connected Vehicle deployment pilots are actively underway in New York, Wyoming, and Tampa Bay [28], with comprehensive documentation of the individual pilot project progress available on the USDOT ITS Portal (https://www.its.dot.gov/pilots/, accessed on 18 July 2023). It is worth noting that there is a relatively limited number of academic journal articles published after 2020 that specifically focus on C-ITS pilots conducted in the United States. Choosakun et al. (2021) and Alanazi et al. (2023) provide comprehensive overviews of C-ITS pilot projects conducted across various countries, including Singapore, Japan, and South Korea [29,30].

In South Korea, C-ITS pilot projects began relatively later compared to Europe. In 2014, the Daejeon–Sejong area was selected as the first C-ITS (Cooperative Intelligent Transport Systems) pilot project in South Korea. Subsequently, as shown in Figure 2, five additional C-ITS service pilot projects were introduced nationwide. Through public bidding processes, the Seoul Metropolitan Government and Jeju Special Self-Governing Province were selected as pilot regions in 2017, followed by the Korea Expressway Corporation’s highway section in 2018, and Gwangju Metropolitan City and Ulsan Metropolitan City in 2019 as experimental areas for C-ITS implementation. Currently, a total of six C-ITS pilot projects are underway (see Figure 2). The Daejeon–Sejong C-ITS pilot project is organized by the Ministry of Land, Infrastructure and Transport (MOLIT), a central government agency. The Seoul-Busan Expressway pilot project is managed by the Korea Expressway Corporation. The remaining pilot projects were initiated through competitive bidding processes and conducted by the relevant metropolitan local governments, with financial support from the central government budget. In Jeju Island, since 2018, rental cars have been equipped with C-ITS devices to provide various hazard warning services to rental car users [31]. The Seoul Metropolitan Government is actively promoting C-ITS implementation and autonomous driving based on public transportation and has made certain data available through the Seoul Transportation Big Data (T-DATA) website (https://t-data.seoul.go.kr/, accessed on 18 July 2023). Over the past decade, various C-ITS pilot projects involving local governments, public agencies, and private developers have fostered technological innovation. The importance of C-ITS as a key technology for driving smart mobility in future cities is increasing day by day. Particularly noteworthy is the remarkable technological growth in the deployment of On-board Units (OBU) in vehicles, expansion of Roadside Units (RSU) facilities, and increased speed of Vehicle-to-Infrastructure (V2I) communication.

In contrast, there is a relative lack of post-analysis and verification regarding the actual effectiveness of the services provided. This is not unique to Korea but is a common phenomenon in all countries implementing C-ITS pilot projects. Most of the related studies have relied on simulation-based research or indirect validation of C-ITS effects through expert or user survey analysis, rather than utilizing actual user data. In particular, the analysis of user driving behavior using the accumulated big data from C-ITS pilot projects rarely exists. Therefore, this study attempts to analyze the driving behavior of service users based on the generated and accumulated operational data from the Daejeon–Sejong C-ITS pilot project. The study encompasses three main objectives: Firstly, it aims to disseminate the preliminary results from the Korean pilot project to support C-ITS researchers worldwide. Secondly, it endeavors to propose a methodology for analyzing travel behavior using C-ITS data. Lastly, the third objective involves testing whether the utilization of C-ITS services has an impact on safe driving practices.

2. Methodology

2.1. Case Study

The analysis in this study focuses on the recorded On-Board Unit (OBU) operational data within the Daejeon–Sejong C-ITS pilot testbed area. It should be noted that these data are recorded solely when vehicles outfitted with C-ITS services operate within the confines of the testbed area, hence the fluctuating count of vehicles availing C-ITS services at different times. The operational data, collected at intervals ranging from 1 to 2 s, is accumulated in the traffic management center.

Daejeon and Sejong in South Korea, with distinct spatial characteristics, serve as significant testbeds for the C-ITS (Cooperative Intelligent Transportation System) pilot study. Daejeon, similar to cities like San Jose in the U.S., with its urban layout at major transport crossroads and status as a technological hub, aligns with its role in C-ITS testing. Sejong, a well-planned administrative city akin to Canberra in Australia, with a controlled urban structure, offers a unique environment for C-ITS experimentation.

Their collaboration, leveraging Daejeon’s technological ecosystem and Sejong’s governmental support, creates a comprehensive testing ground. The spatial and strategic alignment of these cities, combined with the detailed analysis of real-time operational data, sets them apart, making the Daejeon–Sejong corridor vital in the development of intelligent transportation systems. The spatial scope of the experimental area encompasses a comprehensive stretch of 90.7 km of road networks, spanning across Daejeon City and Sejong City. This includes both expressways and regular road segments. A total of 95 Road-Side Units (RSUs) are strategically installed within this experimental area. Figure 3 illustrates a detailed representation of the Daejeon–Sejong C-ITS testbed area, with the green dots representing the locations of RSUs.

The two cities, Daejeon and Sejong, exhibit a property of high scalability in relation to each other, despite their differing sizes. Daejeon, being the fifth-largest metropolis in South Korea, covers an area of approximately 539.86 square kilometers. Sejong, on the other hand, is a smaller, well-planned administrative city, spanning an area of around 465.23 square kilometers.

This scalability reflects their ability to adapt and extend their technological and administrative frameworks, allowing for seamless integration and collaboration. The shared characteristics and complementary roles of these cities in technological innovation and governmental functions create a synergy that enhances their potential for growth and development. This high scalability, transcending the size of the cities, sets them apart and contributes to their significance as models for intelligent transportation systems and smart city initiatives.

Under the supervision of the Ministry of Land, Infrastructure and Transport (MOLIT), efforts were directed towards the recruitment of 3000 citizens, who were vehicle owners hailing from Daejeon City and Sejong City, to constitute the C-ITS experience team. This recruitment process was conducted from June to September 2016. As part of this initiative, C-ITS devices were installed in the vehicles of the participants. In the early stages of the pilot project, the services were provided via Head-up Displays (HUDs) or navigation systems. However, starting from 2017, smartphone app services were introduced. As of 2017, HUD-type vehicle terminals account for approximately 6% of the total, while the share of smartphone-based vehicle terminals escalated to 47.6%. In terms of vehicle types, regular passenger vehicles occupy 95.3% of the total, while police cars, school buses, buses, trucks, and ambulances account for only 4.7%. Thus, it can be reasonably inferred that the majority of users analyzed within this study were operators of standard passenger vehicles outfitted with either smartphone or navigation-based devices.

2.2. Data Description

2.2.1. Extraction of RSU and OBU Data

The essence of Cooperative Intelligent Transportation Systems (C-ITS) is deeply entrenched in the exchange of extensive data sets among diverse entities, inclusive of vehicles, Road-side Units (RSUs), and Traffic Control Centers (TCCs), constructed on multi-layered systems [32]. The importance of parallel and distributed processing for handling large-scale data has been highlighted by Javed et al. (2019) [33]. Prominent engines for effective distributed processing of voluminous datasets encompass Hadoop and Spark. In the context of this study, the Spark engine was harnessed to augment data processing velocity through the deployment of parallel processing methodologies. The execution of distributed processing within the Spark engine can be bifurcated into three sequential stages as delineated in the following.

The preliminary phase of this process involves the retrieval of data from disks and subsequent storage within the framework of the Resilient Distributed Datasets (RDD) data structure. These data are distributed and subsequently stored within the memory of machines that constitute the cluster, thus ensuring reduced I/O overhead given that intermediate results are directly committed to memory, thereby eliminating the need for disk storage, a characteristic feature of the map-reduce approach embodied by Hadoop. The created RDD is characterized by its immutability and exhibits a read-only attribute. In the subsequent stage, transform operators are applied to the RDD. This includes but is not limited to map(), filter(), reduce(), and join(). Instead of yielding instantaneous results, these operators generate new RDDs and defer the computational process. These operations, distributed across the cluster, are executable as individual tasks. The concluding phase entails the use of action operators, which yield tangible results and derive outcomes, represented by operations such as count(), collect(), and saveAsTextFile(). These operators function based on the successful completion of the preceding transform operators. After the analysis and optimization of the pending transform operators, the results are derived accordingly.

The data derived from the Road-Side Units (RSUs) encompass various informational facets, such as transmission time, the spatial positioning of RSUs, and the types as well as codes of warning services.

Table 1. Types and proportions of C-ITS warning services in 2017.

C-ITS Services	Proportion
Dangerous road section ahead warning	42.4%
Traffic information warning	36.1%
Red light signal caution warning	5.8%
School zone warning	5.3%
Weather (temperature and humidity) information	1.8%
Silver zone (for elderly pedestrians) warning	1.4%
Pedestrian crossing warning	1.3%
Signal violation warning	0.9%
Speed bump ahead warning	0.9%
Notification of violating vehicles warning	0.7%
Slow-moving vehicles ahead warning	0.7%
Variable Message Signs (VMS) information	0.3%
Vehicle malfunctions	0.2%
Abandoned vehicle warning	0.1%
Traffic congestion warning	0.1%

represents the distinct types and relative proportions of C-ITS warning services rendered. In 2017, C-ITS services, including warning for hazardous road sections, the provision of traffic information, attention to red signals, and school zone warnings, accounted for 89.6% of the total.

The Daejeon–Sejong C-ITS pilot project encompasses a wide array of warnings and notifications to enhance driver safety and awareness. These include alerts for dangerous road sections, real-time traffic updates, red light cautions, school and elderly pedestrian zones, and pedestrian crossing warnings (Table 1). A school zone is designated around schools, and under the current Korean law, the speed limit for driving is restricted to 30 km per hour (km/h). Silver zones, which are senior protection zones, are designated around areas with a high concentration of elderly people or facilities used by seniors. Each local government imposes speed limits within the range of 30 to 50 km/h in these zones. Weather-related updates, speed bump alerts, and warnings about traffic signal violations, nearby violating vehicles, and slow-moving vehicles are also provided. Additionally, drivers receive information through Variable Message Signs (VMS) and notifications about vehicle malfunctions, abandoned vehicles, and traffic congestion. Together, these services foster a safer driving environment by encouraging caution, timely decision making, and adaptability to varying road conditions and potential hazards (Table 1).

In the case of the Daejeon–Sejong C-ITS pilot project, participation is voluntary, and therefore, there appears to be a high level of participation among owners of regular passenger vehicles. Due to privacy concerns, access to information related to participants is not possible, and there are limitations for researchers in accessing information regarding the trip purpose or final destinations. Each OBU vehicle sends information, such as vehicle type, vehicle location (latitude, longitude), gear state, vehicle speed, emergency light operation status, hard braking status, and emergency light state, to the base station at intervals of 1 to 2 s. As of 2017, more than 2 million data entries were generated on average per day, and the amount of data generated by OBU vehicles has consistently increased up until 2020 (

Table 2. Yearly aggregated OBU data.

	2017	2018	2019	2020
Total amount of data	737,402,453	769,156,169	845,648,138	953,314,750
Year-on-year growth rate	-	4.1%	9.0%	11.3%
Average daily events	2,020,281	2,107,277	2,316,844	2,611,821

).

2.2.2. Variables of Five Traffic Safety-Related Driving Behavior

C-ITS is expected to enhance the safety of vehicles, reduce the causes of congestion, and promote efficient and eco-friendly driving that minimizes carbon emissions [34,35]. Particularly, preventing traffic accidents and promoting safe driving are among the key objectives of C-ITS services. According to the estimation by Rama and Innamaa (2020) [26], various C-ITS warning services have been assessed to reduce the number of “fatality” and “injury crashes”. Weibull et al. (2022) argued that C-ITS has the potential to mitigate the risks of emergency-vehicle-related accidents [36]. In this context, this study focuses on vehicle speed, acceleration, and deceleration as indicators related to traffic safety. Previous studies have demonstrated the correlation between increased vehicle speed and collision risk [37]. Kamrani et al. (2018) claimed that intersection crashes are associated with speed and acceleration [38]. Additionally, Stavrakaki et al. (2020) mentioned evidence linking harsh acceleration and harsh braking to dangerous driving [39].

This study extracted the following five indicators related to safe driving from the OBU data: (1) average speed, (2) hard braking, (3) severe deceleration, (4) speeding, and (5) excessive speeding. Vehicle speed, hard braking events, and severe deceleration events are recorded in the OBU vehicle operation records. Using Azure’s Databricks and the Spark engine for distributed parallel computing, the OBU ID-based average speed, number of hard braking events per month, and number of severe deceleration events per month were calculated. The speeding rate and excessive speeding rate were extracted using a cell-based method, which will be described in detail in the subsequent section.

The temporal scope of the extracted data for analysis is from February 2017, when the C-ITS system began operating in the Daejeon–Sejong C-ITS testbed area. In order to ascertain the influence of C-ITS services in fostering safer driving practices, a comparative analysis leveraging paired-sample t-tests was implemented over a span of twelve months beginning in February 2017. The aggregate count of On-Board Unit (OBU) IDs enlisted for this analysis totaled to 969. For the first two months (February and March 2017) after the completion of OBU installation in the pilot vehicles, only operational data were collected, and C-ITS warning services were not provided. After a period of three months subsequent to the installation of the On-Board Units (OBUs), users were presented with warning services grounded in Cooperative Intelligent Transportation Systems (C-ITS). Through the comparison of behavior during these two periods, we analyze the changes in driving behavior before and after the provision of warning services.

2.2.3. Cell-Based Data Extraction Method

The volume of On-Board Unit (OBU) data amassed within the pilot area under the aegis of the Cooperative Intelligent Transportation Systems (C-ITS) surpasses two million records on a daily basis (refer to Table 2). Despite the theoretical feasibility of querying the desired data within such a voluminous data environment utilizing conventional methodologies, its practical implementation is nearly unattainable. For instance, when extracting the speeding rate based on vehicle speed, the following scenario can be observed. When an OBU-equipped vehicle enters the Daejeon–Sejong pilot project area, it transmits vehicular operation records to the traffic management center at intervals of 1 to 2 s. These vehicle operation records include the timestamp, location, and vehicle speed generated by the OBU vehicle. Based on these data, in the raw data processing, the process of determining speeding and excessive speeding begins by retrieving the speed limit for each road link from the national database. These speed limits are then spatially joined with each corresponding road link, creating a clear connection between the road segments and their legal speed limits. Next, for a given On-Board Unit (OBU) data point, the system calculates the vector distance to all the road links, identifying the link with the minimum distance that is the nearest to the data point. By comparing the speed of the OBU data point with the speed limit of this nearest link, the system calculates the difference between the two. If the vehicle’s speed exceeds the limit, it is flagged for speeding. This entire process is repeated for all OBU data points within the testbed area, allowing for a comprehensive analysis of speeding and excessive speeding. The collected data can then be used to enhance traffic management and road safety, contributing to a more responsible and efficient driving environment. Raw data processing methods yield accurate results, but they suffers from the drawback of significantly increased processing times.

The utilization of a cell-based extraction method in data processing presents a noteworthy advantage in expediting processing time, albeit with a potential compromise in precision. The methodology is delineated as follows: the target area is systematically partitioned into a grid, with cells measuring 300 m by 300 m, as illustrated in Figure 3. Subsequently, an identification number is assigned to each cell, and the vertices’ coordinates are meticulously recorded, thereby providing the cell ID and georeferencing the vertices for each individual cell. The vehicle speed data are then associated with the cell ID, predicated on the coordinates, by joining each cell with the road link data based on proximity, utilizing the coordinates of the end points of each road link. Within this framework, a representative speed for each cell is ascertained, contingent upon the proportion of the total road length encompassed within the cell. The method then computes the discrepancy between the actual vehicle speed and the representative road speed limit for the corresponding cell (Figure 4), in addition to the variance between the speed of the On-Board Unit (OBU) data points within the cell and the representative speed limit of the cell. Operating at the cellular level, this approach realizes markedly superior processing speeds. However, the assignation of a singular representative road speed limit to each cell may engender a certain degree of unavoidable error. Despite this limitation, within a big data context, the cell-based extraction method emerges as a more tenable and efficacious alternative, circumventing the temporal and financial constraints inherent in conventional raw data processing methods. The procedure is reiterated for all cells to procure the final results.

In the cell-based extraction method, determining the optimal cell size that minimizes the error rate while also reducing processing time becomes a critical factor. In this study, the process of determining the optimized cell size for the target area was conducted as follows: Firstly, a sampling area was selected representing the densest road network. Then, within the sampling area, a total of 540,716 OBU data points generated on a specific day were extracted. Spatial join was applied using GIS software (QGIS 3.22) on a local computer, and the speeding rate (the percentage of OBU-installed vehicles exceeding the road speed limit) was calculated. The processing time required to extract the speeding rate from the sample area by using the raw data processing method was 7136 s, equivalent to 1 h and 59 min.

Next, for comparison purposes, the cell-based extraction method was used to extract the speeding rate within the same sampled area. In this case, the cell size was varied from 100 m to 1000 m in increments of 100 m, and the speeding rate was calculated using Databricks. Figure 5 illustrates the relationship between the cell size and the error rate and execution time. As the cell size increases, the error rate increases while the execution time decreases. The accuracy is calculated based on the information allocated for each link in the raw data and is considered as the ground truth. When processing data using a cell-based approach, attribute information (in this study, the speed limit) is assigned to each cell. However, this process may lead to situations where links with different characteristics are assigned to the same cell. Consequently, in cell-based analysis, where each cell is associated with one attribute value per link, the results may differ from the actual values, leading to reduced accuracy as the cell size increases. However, during the computation process, conducting the analysis based on different cell sizes eliminates the need to match link coordinates with links every second, resulting in increased processing speed.

Particularly, as the cell size increases, the number of link coordinate matches decreases, leading to a decrease in computation time. To elaborate on the differences, when processing data using a cell-based approach, attribute information (in this study, the speed limit) is allocated to each cell. However, this process may lead to situations where links with different characteristics are assigned to the same cell. Consequently, in cell-based analysis, where each cell is associated with one attribute value per link, the results may differ from the actual values, leading to reduced accuracy as the cell size increases. However, during the computation process, conducting the analysis based on different cell sizes eliminates the need to match link coordinates with links every second, resulting in increased processing speed. Particularly, as the cell size increases, the number of link coordinate matches decreases, leading to a decrease in computation time.

Comparing the execution times between the raw data processing method and the cell-based extraction method, the latter proves to be significantly more efficient. However, there is a trade-off as larger cell sizes result in higher error rates. Therefore, selecting the most efficient cell size in terms of the error rate versus execution time ratio is crucial. The optimal size is expected to be influenced by the average block size (distance between intersections) and speed limit variation in each region. Based on the results in Figure 5, a cell size of 300 m (with an error rate of 0.1 and an execution time of 34 s) was deemed the most efficient in terms of time. Accordingly, the cell size was set to 300 × 300 m, and the entire area was subjected to cell-based data extraction. Based on cells with 300 × 300 m size, the speeding rate and excessive speeding rate were extracted using the same method, defining speeding as the difference between the actual speed of the OBU-installed vehicle and the road speed limit being below 20 km/h, and defining excessive speeding as exceeding 20 km/h.

In the Daejeon–Sejong C-ITS testbed, the daily average quantity of generated On-Board Unit (OBU) data surpasses 2 million, culminating in an annual accumulation exceeding 700 million data points (refer to Table 2). Consequently, to process and analyze such a voluminous corpus of data within this environment, the employment of the cell-based extraction method and cloud-based parallel processing and distributed computing becomes indispensable. The cell-based extraction methodology implemented within this study, in conjunction with analogous approaches, is anticipated to receive increasingly extensive investigation and adoption within future academic research.

2.3. Analysis

The objective of this study is to validate the short-term effects of C-ITS services by comparing the five safety-related indicators (average speed, hard braking rate, severe deceleration rate, speeding rate, and excessive speeding rate) extracted from C-ITS users based on their OBU IDs. For each indicator, five paired t-tests were conducted. Two-sample t-test is used to compare the means of two independent samples, which is the most fundamental statistical test. On the other hand, when the observations between two samples are not independent and are measured in pairs with mutual influence, the paired t-test is employed to examine the differences between them [40]. For example, when the same sample is measured before and after an intervention or when there is dependency due to sibling or twin relationships, the paired t-test is utilized [40]. Two-sample t-test assumes that the data from both samples follow a normal distribution and have equal variances, whereas the paired t-test only requires that the differences between pairs follow a normal distribution [40].

Paired t-tests are frequently used in medical research, which often involves pre–post comparisons within the same group, but they are relatively less utilized in transportation and urban planning research, where it is often challenging to conduct longitudinal studies on the same group. In relevant studies, Aljoufie (2021) employed a paired t-test to investigate the temporal impact of population density regulations in Jeddah, Saudi Arabia, by combining population density variables and temporal impact indicators based on data from 2007 and 2014 [41]. Arora and Kumar (2021) utilized a paired t-test to verify if there was a significant difference in the quality between India’s taxi-aggregator service and traditional public transportation service, based on a survey on service quality [42]. Hoj et al. (2018) used a paired t-test to compare the pre- and post-usage of electric bicycles (e-bikes) by measuring users’ rides with smartwatches and conducting surveys [43]. Li et al. (2022) conducted a case study based on Beijing’s traffic congestion index data and compared the differences between two time points with a one-year interval using a paired t-test [25]. Transportation big data is accumulated over a long period based on user IDs, providing optimized data for within-group comparisons, which is expected to increase the use of paired t-tests in future transportation and urban planning research.

The paired t-test is used when two sample groups are paired, comparing the difference in means between the two groups. It is employed in cases where the same individuals are measured before and after a certain intervention or exposure. Specifically, it is used to compare a pre-exposure group (Group A) with the same individuals after exposure (Group B) to a specific service or treatment. The paired t-test is defined as follows:

$$D_{i } = X{ }_{\mathrm{Ai}} - X_{\mathrm{Bi}} (i= \mathrm{1,2},3, \dots n)$$

The t-test statistic for paired sample hypothesis testing is calculated as follows, assuming the normality of the sample data: In cases where the sample size is small (n < 30), the Shapiro–Wilk’s W statistic is used to assess the normality of the data, which yields a value between 0 and 1. The t-test statistic can be computed as follows:

$$t= \frac{\overline{D}}{S\sqrt{n}} (S_{\mathrm{ij}}=\frac{\sum _{i=1}^n(D_i-\overline{D}{)}^2}{n-1}, \overline{D} = \frac{1}{n}\sum _{i=1}^n{D}_i)$$

In the initial phase of the Daejeon–Sejong C-ITS pilot project, which took place in February and March 2017, no warning services were provided. Therefore, in this study, the OBU vehicle operation data during this period is considered as data prior to exposure to C-ITS services. The analysis was conducted on a total of 969 OBU vehicles collected over a period of 12 months from February 2017 to January 2018, to examine the short-term effects of C-ITS services on driving behavior, specifically focusing on average speed, hard braking rate, severe deceleration rate, speeding rate, and excessive speeding rate. Paired sample t-tests were performed with a significance level of 0.05 to investigate whether there were any differences in the driving behavior of users before and after receiving C-ITS services. The alternative hypothesis for the paired sample t-test is as follows:

$$H_{Ai} : {\mu }_{ai }\ne {\mu }_{bi }$$

where:

HAi: Alternative hypothesis (i.e., there are changes in average speed, hard braking rate, severe deceleration rate, speeding rate, and excessive speeding rate);

μai: Average speed, hard braking rate, severe deceleration rate, speeding rate, and excessive speeding rate of the 969 OBU vehicles during the period without receiving C-ITS services (February 2017);

μbi: Average speed, hard braking rate, severe deceleration rate, speeding rate, and excessive speeding rate of the 969 OBU vehicles during the period receiving C-ITS services (April 2017 to January 2018).

3. Results

The analysis of driving behavior among C-ITS service users through paired sample t-tests yielded the following results.

3.1. Average Speed

During the period of receiving C-ITS warning services, it can be observed that the average driving speed gradually decreased compared to the average speed in February 2017, which represents the period before receiving C-ITS services. Particularly, after June 2017, the average speed of vehicles remained consistently lower than the average speed in February 2017 (except for temporary increases in May and year-end periods) (see Figure 6).

Table 3. Result of the paired t-test: average speed.

		Average	S.D	t	p
paired 1	Feb-2017	41.00	15.79	0.42	0.68
	Mar-2017	40.85	14.81
paired 2	Feb-2017	41.00	15.79	1.42	0.15
	Apr-2017	40.47	14.46
paired 3	Feb-2017	41.00	15.79	−3.64	0.00 *
	May-2017	42.44	15.46
paired 4	Feb-2017	41.00	15.79	3.58	0.00 **
	Jun-2017	39.54	14.50
paired 5	Feb-2017	41.00	15.79	6.83	0.00 **
	Jul-2017	38.24	14.95
paired 6	Feb-2017	41.00	15.79	6.34	0.00 **
	Aug-2017	38.35	14.58
paired 7	Feb-2017	41.00	15.79	9.19	0.00 **
	Sep-2017	37.35	14.42
paired 8	Feb-2017	41.00	15.79	8.55	0.00 **
	Oct-2017	37.46	13.67
paired 9	Feb-2017	41.00	15.79	9.69	0.00 **
	Nov-2017	37.21	13.91
paired 10	Feb-2017	41.00	15.79	7.53	0.00 **
	Dec-2017	38.01	14.06
paired 11	Feb-2017	41.00	15.79	5.72	0.00 **
	Jan-2017	38.48	15.78

*: At a significance level of 0.01, it is statistically significant that the average speed in the compared month (May 2017) is higher than the average speed in the reference month (February 2017). **: At a significance level of 0.01, it is statistically significant that the average speed in the compared month is lower than the average speed in the reference month (February 2017).

presents the results of paired sample t-tests regarding the differences in monthly average speed compared to February 2017. The p-values for average speed from June 2017 to January 2018 were found to be smaller than the significance level of 0.01, leading to the rejection of the null hypothesis and the acceptance of the alternative hypothesis for the period. Based on these results, it can be inferred that the average speed of vehicles after a certain point is statistically significantly lower compared to the period without such services. In this case, this change in average speed did not immediately occur after the initiation of the services and showed a time lag of approximately three months.

3.2. Hard Braking Rate

Figure 7 illustrates the monthly variations in the hard braking rate. The hard braking rate during the period of receiving C-ITS services shows a decrease compared to the rate before receiving the services. The decrease in the hard braking rate is observed immediately after the provision of C-ITS services, without any apparent time lag.

Table 4. Result of the paired t-test: hard braking rate.

		Average	S.D	t	p
paired 1	Feb-2017	0.08	0.19	−2.35	0.02
	Mar-2017	0.09	0.20
paired 2	Feb-2017	0.08	0.19	3.38	0.00 **
	Apr-2017	0.07	0.18
paired 3	Feb-2017	0.08	0.19	5.43	0.00 **
	May-2017	0.06	0.17
paired 4	Feb-2017	0.08	0.19	5.57	0.00 **
	Jun-2017	0.06	0.17
paired 5	Feb-2017	0.08	0.19	6.32	0.00 **
	Jull-2017	0.05	0.16
paired 6	Feb-2017	0.08	0.19	6.09	0.00 **
	Aug-2017	0.05	0.16
paired 7	Feb-2017	0.08	0.19	6.70	0.00 **
	Sep-2017	0.05	0.16
paired 8	Feb-2017	0.08	0.19	6.21	0.00 **
	Oct-2017	0.05	0.16
paired 9	Feb-2017	0.08	0.19	6.81	0.00 **
	Nov-2017	0.05	0.16
paired 10	Feb-2017	0.08	0.19	7.57	0.00 **
	Dec-2017	0.05	0.15
paired 11	Feb-2017	0.08	0.19	5.67	0.00 **
	Jan-2017	0.05	0.17

**: At a significance level of 0.01, it is statistically significant that the hard braking rate in the compared month is lower than the hard braking rate in the reference month (February 2017).

presents the results of paired sample t-tests regarding the differences in monthly hard braking rate compared to February 2017. The p-values for the hard braking rate from April 2017 to January 2018 were found to be smaller than the significance level of 0.01, leading to the rejection of the null hypothesis and the acceptance of the alternative hypothesis. Overall, the hard braking rate during the period when C-ITS services were employed is relatively lower compared to February 2017, and this difference is statistically significant.

3.3. Severe Deceleration Rate

Figure 8 displays the monthly trend of the severe deceleration rate, which shows a relative decrease compared to February 2017. Similar to the hard braking rate, the severe deceleration rate also exhibits a decrease immediately after the provision of C-ITS warning services, without any apparent time lag.

Table 5. Result of the paired t-test: severe deceleration rate.

		Average	S.D	t	p
paired 1	Feb-2017	0.17	0.31	−3.21	0.00 *
	Mar-2017	0.19	0.33
paired 2	Feb-2017	0.17	0.31	2.78	0.01 **
	Apr-2017	0.16	0.29
paired 3	Feb-2017	0.17	0.31	8.64	0.00 **
	May-2017	0.11	0.24
paired 4	Feb-2017	0.17	0.31	9.85	0.00 **
	Jun-2017	0.10	0.23
paired 5	Feb-2017	0.17	0.31	10.11	0.00 **
	Jul-2017	0.10	0.23
paired 6	Feb-2017	0.17	0.31	8.87	0.00 **
	Aug-2017	0.11	0.24
paired 7	Feb-2017	0.17	0.31	11.77	0.00 **
	Sep-2017	0.08	0.19
paired 8	Feb-2017	0.17	0.31	10.14	0.00 **
	Oct-2017	0.10	0.23
paired 9	Feb-2017	0.17	0.31	11.54	0.00 **
	Nov-2017	0.09	0.21
paired 10	Feb-2017	0.17	0.31	10.93	0.00 **
	Dec-2017	0.09	0.22
paired 11	Feb-2017	0.17	0.31	11.50	0.00 **
	Jan-2017	0.08	0.21

*: At a significance level of 0.01, it is statistically significant that the Severe Deceleration Rate in the compared month (March 2017) is higher than the Severe Deceleration Rate in the reference month (February 2017). **: At a significance level of 0.01, it is statistically significant that the Severe Deceleration Rate in the compared month is lower than the Severe Deceleration Rate in the reference month (February 2017).

presents the results of paired sample t-tests regarding the differences in monthly severe deceleration rate compared to February 2017. The p-values for the severe deceleration rate from April 2017 to January 2018 were found to be smaller than the significance level of 0.01, leading to the rejection of the null hypothesis and the acceptance of the alternative hypothesis. This indicates that the severe deceleration rate for users after receiving C-ITS services is statistically significantly lower than that of February 2017 when C-ITS warning services were not provided. Both hard braking and severe deceleration are considered as prominent risky driving behaviors, and the decrease in the rates of these two indicators suggests a possible shift towards safer driving behavior among users after receiving C-ITS services.

3.4. Speeding Rate

Figure 9 illustrates the monthly trend of the speeding rate. The speeding rate shows a gradual decrease after receiving C-ITS services. Unlike the previous indicators, the effect on the speeding rate was not immediately apparent after the service provision and took around four months to manifest.

Based on the results of the paired sample t-test shown in

Table 6. Result of the paired t-test: speeding rate.

		Average	S.D	t	p
paired 1	Feb-2017	0.04	0.05	−1.69	0.09
	Mar-2017	0.04	0.05
paired 2	Feb-2017	0.04	0.05	−0.78	0.44
	Apr-2017	0.04	0.04
paired 3	Feb-2017	0.04	0.05	−1.77	0.08
	May-2017	0.04	0.05
paired 4	Feb-2017	0.04	0.05	0.71	0.48
	June-2017	0.04	0.04
paired 5	Feb-2017	0.04	0.05	4.79	0.00 **
	Jul-2017	0.03	0.04
paired 6	Feb-2017	0.04	0.05	7.06	0.00 **
	Aug-2017	0.03	0.04
paired 7	Feb-2017	0.04	0.05	6.37	0.00 **
	Sep-2017	0.03	0.04
paired 8	Feb-2017	0.04	0.05	4.32	0.00 **
	Oct-2017	0.03	0.04
paired 9	Feb-2017	0.04	0.05	5.34	0.00 **
	Nov-2017	0.03	0.04
paired 10	Feb-2017	0.04	0.05	6.88	0.00 **
	Dec-2017	0.03	0.04
paired 11	Feb-2017	0.04	0.05	8.97	0.00 **
	Jan-2017	0.03	0.04

**: At a significance level of 0.01, it is statistically significant that the Speeding Rate in the compared month is lower than the Speeding Rate in the reference month (February 2017).

, the decrease in the speeding rate becomes statistically significant from July 2017 onwards, as indicated by p-values smaller than the significance level of 0.01. Thus, it can be concluded that the speeding rate for users after receiving C-ITS services is lower compared to February 2017 when C-ITS services were not provided.

3.5. Excessive Speeding Rate

Figure 10 depicts the monthly trend of the excessive speeding rate during the analyzed period. Overall, the excessive speeding rate during the period of receiving C-ITS services shows a gradual decrease compared to February 2017 when warning services were not provided. Similar to the speeding rate, the effect on the excessive speeding rate was not immediately observed after the service provision and showed a time lag of approximately four months.

Table 7. Result of the paired t-test: excessive speeding rate.

		Average	S.D	t	p
paired 1	Feb-2017	0.04	0.05	−1.69	0.09
	Mar-2017	0.04	0.05
paired 2	Feb-2017	0.04	0.05	−0.78	0.44
	Apr-2017	0.04	0.04
paired 3	Feb-2017	0.04	0.05	−1.77	0.08
	May-2017	0.04	0.05
paired 4	Feb-2017	0.04	0.05	0.71	0.48
	Jun-2017	0.04	0.04
paired 5	Feb-2017	0.04	0.05	4.79	0.00 **
	Jul-2017	0.03	0.04
paired 6	Feb-2017	0.04	0.05	7.06	0.00 **
	Aug-2017	0.03	0.04
paired 7	Feb-2017	0.04	0.05	6.37	0.00 **
	Sep-2017	0.03	0.04
paired 8	Feb-2017	0.04	0.05	4.32	0.00 **
	Oct-2017	0.03	0.04
paired 9	Feb-2017	0.04	0.05	5.34	0.00 **
	Nov-2017	0.03	0.04
paired 10	Feb-2017	0.04	0.05	6.88	0.00 **
	Dec-2017	0.03	0.04
paired 11	Feb-2017	0.04	0.05	8.97	0.00 **
	Jan-2017	0.03	0.04

**: At a significance level of 0.01, it is statistically significant that the Excessive Speeding Rate in the compared month is lower than the Excessive Speeding Rate in the reference month (February 2017).

presents the results of the paired sample t-test for the monthly excessive speeding rate. From July 2017 onwards, the p-values for the excessive speeding rate are smaller than the significance level of 0.01, leading to the rejection of the null hypothesis and the acceptance of the alternative hypothesis. Therefore, it can be concluded that the excessive speeding rate during the period of receiving C-ITS services is lower compared to February 2017 when C-ITS services were not provided. The analysis of both the speeding rate and excessive speeding rate in this study indicates a decreasing trend, suggesting that the driving behavior of users has shifted towards safer practices after receiving C-ITS services, assuming that both factors hinder safe driving.

The present study aimed to verify the changes in driving behavior based on the provision of C-ITS services. Monthly data from February 2017 to January 2018 were collected from 969 vehicles equipped with On-Board Units (OBU) in the Daejeon–Sejong pilot area. Paired sample t-tests were conducted to compare the data between the period of receiving C-ITS services (April 2017 to January 2018) and the pre-service period represented by February 2017. The analyzed metrics included average speed, hard braking rate, severe deceleration rate, speeding rate, and excessive speeding rate. The analysis revealed that C-ITS users demonstrated safer driving behavior during the period of receiving C-ITS services compared to the pre-service period. However, the timing of the positive effects on each metric varied.

The average speed exhibited a general decreasing trend; however, there were certain points (February and May) where it showed an increasing trend. On the other hand, hard braking and severe deceleration rates demonstrated an immediate decrease following the provision of C-ITS services. The average speed exhibited changes after a lapse of three months, while the speeding rate and excessive speeding rate showed changes four months after the provision of C-ITS services. The order of observed changes in driving behavior was hard braking and severe deceleration rates > average speed > speeding and excessive speeding rates. The results of the paired sample t-tests suggest that, assuming other control factors remain constant, users’ driving behavior gradually stabilizes and becomes safer over the 12-month period following the provision of C-ITS services. This interpretation aligns with the notion of “driving behavior stabilization” discussed by Stavrakaki et al. (2020) [39] while also suggesting the potential existence of specific time points for the stabilization of driving behavior for each metric.

4. Discussion and Conclusions

4.1. Implications

The rapid development of technology and the execution of pilot projects related to Connected Vehicles are phenomena observed worldwide. In Korea, six pilot or demonstration projects related to Connected Intelligent Transport Systems (C-ITS) are currently underway. However, research that aims to demonstrate the effectiveness of these services is still in its infancy. In particular, there is a scarcity of analyses based on actual driving data generated by C-ITS services.

The analysis of large-scale accumulated big data, a byproduct of C-ITS initiatives, necessitates the development of new data extraction methods. These methods must be based on parallel distributed computing approaches to handle the volume and complexity of the data. In response to this need, this study developed a novel process for extracting variables using parallel processing techniques. This was achieved through the use of the Spark engine in DataBricks, a cloud computing platform. The driving data collected from the Daejeon–Sejong C-ITS pilot project served as the basis for this process. The process was applied to cell-based data extraction, and the hypothesis that there may be positive changes in users’ driving behavior after the provision of C-ITS services was examined.

For the first two months after the completion of On-Board Unit (OBU) installation in the pilot vehicles in February 2017, only driving data were collected, and C-ITS warning services were not provided. This study conducted a paired-sample t-test to verify the short-term effects over the subsequent one-year period starting from February 2017. The analysis results showed that, overall, drivers’ driving behavior in terms of average speed, hard braking rate, severe deceleration rate, speeding rate, and excessive speeding rate changed in a safer direction compared to the period before receiving C-ITS warning services.

While there were immediate effects observed for hard braking and severe deceleration rates right after the provision of services, the reduction in the excessive speeding rate was statistically significant after a four-month period following the provision of services. This suggests that C-ITS service users may require some time for their driving behavior to change. The verification of service effectiveness demands further research on other control factors that may influence safety-related driving behavior.

The results of this study underscore the potential benefits of widespread deployment of C-ITS services. The observed improvements in driving behavior following the provision of C-ITS services indicate that such services can play a crucial role in enhancing road safety. As such, these findings provide a compelling case for the broader implementation of C-ITS services. However, the deployment of these services should not be viewed as a one-off intervention, but rather as a long-term strategy that requires continuous monitoring and adjustment.

The study also highlights the importance of long-term monitoring of C-ITS services. The significant reduction in excessive speeding was observed only after four months of service provision, suggesting that changes in driving behavior may take time to manifest. This underscores the need for long-term planning and monitoring in the deployment of C-ITS services. Furthermore, user acceptance of C-ITS services may evolve over time, influenced by factors such as changes in driving behavior, perceived benefits of the services, and technological advancements. Therefore, continuous monitoring and evaluation are essential to ensure that C-ITS services remain effective and relevant, and to facilitate necessary adjustments in response to changing circumstances. This long-term perspective will be crucial in maximizing the potential benefits of C-ITS services and ensuring their successful integration into the transportation system.

Future quantitative analysis of C-ITS service users based on the actual utilization of various warning and information services provided through the on-board OBU terminal will be a crucial step for the objective evaluation and improvement of C-ITS systems. Ultimately, the results of the effectiveness verification will significantly contribute to enhancing public acceptance of C-ITS service-based cooperative autonomous driving policies in the future.

One of the challenges in C-ITS research lies in data accessibility. While autonomous driving big data generated by private companies is often monopolized by them, the collection and management of C-ITS data fall under the jurisdiction of government agencies. Due to the sensitive nature of personal data protection, researchers’ access to C-ITS data is limited. Despite these limitations, with the advancement of Connected Automated Vehicle (CAV) technology, the C-ITS penetration rate is expected to increase gradually, leading to a higher volume of collected data and approaching comprehensive surveys.

The results of this study, particularly the positive impact of C-ITS service on driving behavior, carry significant global implications. The observed safer shift in driving behavior following the provision of C-ITS services in the Korean context suggests a potential for similar outcomes in other regions. This global applicability amplifies the importance of this study, potentially contributing to road safety improvements worldwide.

4.2. Limitations and Future Research

The current study has limitations due to the inherent constraints of the available OBU data. Firstly, the analysis was conducted based on OBU ID, but if multiple drivers operated the same OBU-equipped vehicle, it is limited to consider the results as behavioral changes of the same driver. Secondly, with the analyzed OBU data in this study, it is not possible to determine whether users voluntarily discontinued their usage. For example, if individuals change jobs and no longer need to commute through the pilot area, their data are no longer recorded. However, with the current data, it is not possible to separate such cases from voluntary withdrawal or device malfunction. Thirdly, among the vehicles operating in the pilot area, only a small number receive C-ITS services, so there may be influences on driving behavior from other vehicles, which is a limitation to consider. Moreover, the Daejeon–Sejong pilot zone includes both highways and regular roads, which are different in terms of traffic flow. The difference may cause another limitation of the study.

In the future, by incorporating various social and economic data related to C-ITS users, more comprehensive and integrated modeling can be achieved, enabling more detailed analysis. To promote safe driving through C-ITS services and contribute to the realization of cooperative autonomous driving, the long-term analysis of users’ driving behavior is necessary. Based on such analyses, customized services tailored to users with different characteristics can be possible to provide. Additionally, by adding accident data and road environment variables, along with social and economic variables of C-ITS service users, a more detailed and comprehensive analysis of driving behavior can be conducted. This will allow for a more precise validation of the effectiveness of C-ITS.