GIS has been utilized in transportation safety application. It can analyze and process data to support decision making. Therefore, many research studies applied GIS to investigate spatial pattern analysis, using Moran’s I statistic which creates the understanding of accident phenomenon in the cities. Anselin [11] and Black [12] explained that the principles of spatial pattern analysis are that this approach to network autocorrelation analysis are mathematically equivalent to spatial autocorrelation analysis. Additionally, it focuses on understanding phenomena occurring specifically in highly urbanized areas. Map display and Moran’s I statistic can help researchers to predict situations accurately. Rukewe et al. [13] investigated and documented the variations in crash frequencies by types and across different road types in Ibadan, Nigeria, using GIS as a tool for research. The results of using Moran’s I statistic showed that crashes were clustered in space with motorbike crashes showing greatest clustering. This created the understanding of variations in road traffic accident occurrence as well as identifying locations and neighborhoods with unusually higher accidents frequency. Truong and Somenahalli [14] studied spatial autocorrelation analysis of pedestrian-vehicle crash data for identification and ranking of unsafe bus stops. The spatial autocorrelation tool was used to compute Moran's I statistics and z-scores. The results showed that the z-scores reached the highest value of 136.2, meaning that the pedestrian-vehicle crash data was clustered until a distance threshold of 1,000 m. Mohaymany et al. [15] explored the distribution of road accident occurrence on Arak-Khomein rural road in Markazi province in Iran, using GIS through Moran’s I and Kernel Density Estimation (KDE). The results of using Moran’s I showed that Moran’s index was between 0 to -1 during 2006-2008, showing that the locations of accident occurrence were unclearly scattered (Moran’s index 0.20, z-score 1.18). The use of KDE could identify accident locations more accurately and could be compared with the statistical value of accidents.

Many research studies applied GIS to investigate road accident-prone areas. Levine et al. [16], Affum and Taylor [17], Austin et al. [18], Kim and Levine [19] and Miller [20] improved pin maps of crash and database queries by using GIS. To demonstrate spatial crash distribution at the road network level, other effective analytical tools in GIS software including buffer, nearest neighbor method, simple density and KDE method of crash cluster detection were combined with GIS. Pulugurtha et al. [21] Pulugurtha and Pasupuleti [22] applied the GIS-based methodology to identify high pedestrian crash zones, consisting of geocoding crash data, and creating crash concentration maps and identifying high pedestrian crash zones in Las Vegas metropolitan area and Charlotte, North Carolina as the study area. The results showed that GIS-based methodology reduced the subjectivity in the analysis process and could make the analysis more scientifically reliable. Similarly, Schneider et al. [23] explained that utilizing GIS in the research could reduce problems of accident locations and help formulating preventive plans effectively. Ziari and Khabiri [24], Steenberghen et al. [25] designed pedestrian and bicycle lanes by applying the results from GIS analysis to generate a contour map that identified areas of high crash occurrence determined by crash density and clusters of crashes involving pedestrians and bicyclists. Anderson [26] identified road accident hotspots in London. The study applied the methodology using GIS and KDE to study the spatial patterns of injury-related road accidents. Rankavat and Tiwari [27] identified pedestrian accident-prone locations in Delhi using GIS and KDE. Results showed that fatal pedestrian accidents were significantly clustered (p<0.05) over the Ring Road of Delhi at intersections.

According to the previous studies, GIS has been widely used in the crash analysis to identify high-risk neighborhoods, areas of vehicle collision and accident black spots. GIS provides tools and techniques for identifying and analyzing the influence of location on the phenomenon. It is also a tool which assists planners in making the decision using statistical principles to test. The use of Moran’s I statistic is suitable for making understanding on an overview of accident distribution displayed as a statistical value. KDE shows an analysis of road accident-prone areas with obvious locations. However, previous research studies focus on road network analysis and did not address specifically school zones which are important places in cities. Therefore, this is a key principle has been emphasized by the researchers.

• The digital map of Bangkok, was provided by the City Planning Department of the Bangkok Metropolitan (CPD) and was imported in the data frame of the Esri’s ArcGIS software
• The pedestrian-vehicle crash data in 2016 was collected from Thai Road Accidents Data Center for Road Safety Culture (ThaiRSC) (accident data was recorded from road accident information in compulsory insurance claims data from the Road Accident Victims Protection Co., Ltd. Nationwide). The data were recorded in an Excel spreadsheet and was imported into GIS by giving the X and Y coordinates of each accident. The results of this data were in the Shapefile format (.shp).
• Further analysis was done using SQL (structured query language) in the attribute table of accident points and spatial analysis using spatial statistics tool.

Fig. (1). The research methodology.

Geocoding is the process of assigning a location, usually in the form of coordinate values to an address by comparing the descriptive location to the reference material. With addresses, address locations are spatially displayed which helps in recognizing patterns within the geocoded information. The crash database with location details is imported into GIS. The pedestrian-vehicle crashes are then overlaid on to the road network map. After the data is recorded on GIS, recording attribute data of accident location including fatal, injury, fatal per accidents index has proceeded so as to use in analyzing a relationship of the accident.

Bangkok was selected as the study area because Bangkok metropolis has the highest road accident rate in Thailand (rate of 28,537 cases, or 10% of the entire country). Bangkok is a center of administration, economy, education and employment of the country and then has more than 15 million populations with more than 7.2 million vehicles. In addition, it tends to continually grow at the rate of 2.3 percent per year. There are 748 pedestrian accident cases or 2.47 percent of the entire accidents. The accident rate is likely to increase. This data demonstrates the road accident problem in Bangkok which should be solved immediately [2]. The sample of this study included 436 schools in both primary and secondary schools levels in 50 Bangkok Metropolis Administration (BMA) areas. However, since the study included 436 schools, the researcher addressed only the school zones with severe problems to identify patterns and locations of accident occurrence. The random sampling was distributed across six urban districts. The schools with the highest accident rate in each district were selected to describe phenomena of accident occurrence. GIS was applied as a tool to study spatial pattern analysis and identify pedestrian crash zones.

A multi-stage sampling technique was applied to select the sample, using the highest accident statistics as a filter. The sample was screened from 436 schools across Bangkok and schools with severe accidents were selected as representatives of the six districts. In summary, there were 12 representative schools including six primary schools and six secondary schools in Bangkok. The sampling began with:

• Dividing schools into six district groups
• Administrative areas according to the BMA, consisting of six district groups, were identified. The six areas with the highest rates of vehicle and pedestrian accidents were then selected.
• Selecting districts in Bangkok
• The district representatives of each district group, based on the ThaiRSC statistics of pedestrian-vehicle crashes data in 2016, were selected. There were six selected districts including Bang Sue District, Lat Phrao District, Prawet District, Huai Khwang District, Bang Khun Thian District, and Nong Khaem District Fig. (2).
• Selecting the schools

Fig. (2). Distribution of samples in Bangkok. (Resource: ThaiRSC data applied in GIS by the authors).

The schools were selected from the six district groups. Taking the ThaiRSC statistics on walking distance for Thai people into consideration, these were within 800 m of the schools [28]. Such district groups were the representatives of the school zones where students could access through walking. Such areas were sensitive areas and could cause loss to students when road accidents occurred. Then, the areas with the highest road accidents at each study site were selected. To summarize, 12 schools from the six district groups were selected and the selected schools were analyzed pedestrian crash zones (Table 1).

In this step, the 1,218 crash locations were recorded into the GIS and attributed data of accident locations consisting of fatal, injury, fatal per accidents index was recorded. Spatial autocorrelation analysis was applied to detect the spatial patterns of pedestrian crashes. The statistical calculation of the distance rates of the severity point between schools were used in the analyses. Spatial autocorrelation analysis was used to measure and test how severity points were clustered / dispersed in space with respect to their distances rate. The benefits to create understanding of specific patterns of the phenomenon of accidents, and compare statistical figures which explain the results. Fig. (3) shows the spatial distribution of pedestrian crashes within 800 m of 12 schools (six primary schools and six secondary schools). For the primary schools level, Watudomrungsee School had the highest accident rate (153 cases). For the secondary schools level, Triamudomsuksa Pattanakarn Ratchada School had the highest accident rate (156 cases). However, it is very difficult to identify locations that had multiple crashes using this figure. In order to investigate the spatial dependency among crash frequencies and distances from the school, Moran’s I is a more prevalent technique defined below as ranging from -1 to +1. Moran’s I indicator was employed for the crashes.

Fig. (3). The spatial distribution of pedestrian-vehicle crashes within 800 m of schools, 2016.

The results of the calculation of the autocorrelation statistics of pedestrian crash zones are summarized in Table 2 Based on the spatial pattern analysis through the use of Moran’s I processed by GIS, in the primary schools level, Bangkuntiansuksa School had the lowest value 0.518 while Pracharatbamphen School had the highest value i.e. 0.977. In the secondary school level, Triamudomsuksapattanakarn Ratchada School had the lowest value; 0.532 while Rattanakosin Somphod Bangkhunthian School had the highest value; 0.904. When considering Moran’s I values and z-scores, the results illustrated that locations of accident occurrence among 12 schools showed clustered patterns between 0 to +1. Moreover, the clustered patterns tended to approach +1. When considering z-scores, 12 schools had z-scores greater than 2.58. This showed that accident distribution showed the clustered and certain patterns at statistically significant of 0.01. However, the statistical data showed the tests of the relationship between locations of the accident and school zones only which creates the understanding of phenomena of accident occurrence and comparisons among schools and to relate to environment of school zones to explain the results.

Fig. (4). Pedestrian crash zones within 800 m of schools, 2016.

The analysis of pedestrian crash zones employed the statistics of road accidents within 800 m of the schools in 2016. The Kernel’s distribution patterns were different based on the environment in each area. Once-high density zones are defined, pedestrian safety programs can be focused on them with increased efficiency [36]. The pedestrian crash zones map is shown in Fig. (4), with five crash concentration levels (Very Low, Low, Medium, High, and Very High).

This step focuses on summarizing high pedestrian crash zones for allocation of resources. Table 3. shows the crash frequency of accidents occurrence, fatal accidents, injuries, fatal per accidents index and high crash zones. The analytical results can be categorized based on the levels of education as follows.

4.2.2. Secondary School Level

The school zone of Rattanakosinsomphod Bangkhunthian School was identified as the highest crash zone in the secondary schools level covering 3,370 Sq.m. The locations of accident-prone areas were clustered at the intersection of Rama II Road and Rama II 69 Lane as well as area along the road to the front of the school. This is because there was heavy traffic throughout the day and many mixed land use activities were found in this area. In addition, a large departmental store, fresh market and a local bus stop are located in this area, which is similar to the school zone of Triamudomsuksapattanakarn Ratchada School which was identified as the second highest crash zone covering 2,301 Sq.m. The distribution of accident-prone areas occurred evidently along Rachadaphisek Road, especially at the intersections of main roads. These include the intersections of Ratchadaphisek Road and Ratchadaphisek 7 Lane, Ratchadaphisek Road and Prachasongkhro Road and Pracharajbampen and Ratchadaphisek Road and Ratchdaphisek 3 Lane. This area has large departmental stores, hotels and residential buildings. Accident frequency and severity was also found there. In the secondary schools level, the school zone of Triamudomsuksapattanakarn Ratchada School was identified to have the highest accident rate at 156 cases per year. When considering the accident severity, Ratdamri School was identified to have the highest rate of severe accidents with 2 fatal accidents per year. The dead persons per accidents Index was 0.077.

Fig. (5). Accident-prone areas in school zones in primary schools level (marked by black circles).

In summary, the KDE of 12 schools from six district groups showed that Rattanakosinsomphod Bangkhunthian School, Triamudomsuksapattanakarn Ratchada School, Bangkuntiansuksa School, Wat Lat Phrao School and Yothinborana 2 School were in the most accident-prone areas. The accident locations were different depending on the environment. Importantly, the intersections of arterial roads and local roads, and commercial areas were the most significant areas of vehicle and pedestrian accidents Figs. (5-6) showed the accident-prone areas, that are marked using black circles). The results can be contributed to the government agencies to solve problems at accident-prone areas. Moreover, the results support decision making for transportation planners to create solutions and preventive measures for students. However, such accidents are a result of several factors such as the road and physical environment, vehicle and driver behavior and pedestrian behavior. This will be investigated in the further study.