See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/291973317

Spatial and statistical analysis of road accidents hot spots using GIS

Conference Paper · December 2015

CITATIONS READS
24 8,446

3 authors:

Jayvant Choudhary Anurag Ohri

Madhav Institute of Technology & Science Gwalior Indian Institute of Technology (Banaras Hindu University) Varanasi
30 PUBLICATIONS   284 CITATIONS    71 PUBLICATIONS   446 CITATIONS   

SEE PROFILE SEE PROFILE

Brind Kumar
Indian Institute of Technology (Banaras Hindu University) Varanasi
45 PUBLICATIONS   609 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Thermal monitoring of Ganga river View project

Utilization of Waste Materials as Alternative Fillers in Bituminous Mixes View project

All content following this page was uploaded by Jayvant Choudhary on 27 January 2016.

The user has requested enhancement of the downloaded file.

 3rd Conference of Transportation Research Group of India (3rd CTRG)

Spatial and statistical analysis of road accidents hot spots

using GIS
Jayvant Choudharya,*, Anurag Ohrib, Brind Kumarc
a
Research Scholar, Indian Institute of Technology (Banaras Hindu University), Varanasi-221005 , India,
[email protected]
b
Assistant Professor , Indian Institute of Technology (Banaras Hindu University), Varanasi-221005 , India, [email protected]
c
Assistant Professor , Indian Institute of Technology (Banaras Hindu University), Varanasi-221005 , India,
[email protected]
* Corresponding Author

Abstract. The prime goal of highway safety engineering is to limit the number as well as severity of
traffic accidents by identifying, implementing and evaluating measures to improve highway safety. As
the roadway improvements are supposed to be applied to hazardous locations or accident hot spots where
they have the most momentous impact, identification of hot spots is a vital step in safety management.
There is an ongoing determination by road professional and academicians alike to examine most
appropriate methods for identifying road accidents hot spots. There is no universal definition of hot spot
and it is open to wide speculation. This research seeks to take advantage of GIS and spatial analysis to
identify road accident hot spots both visually and statistically.

Spatial clustering of accidents and spatial densities of hot spots were evaluated using Moran’s I method
of incremental spatial autocorrelation, Getis-Ord Gi* statistics and Kernel Density Estimation (K). The
statistical techniques were compared using five years of accident data (2009-2013) for Varanasi city.
Apart from accident counts, severity indices were also employed for analysis and ranking of hot spots
using both methods. This approach employed different severity weighing systems to evaluate the effect
of Minor and Property Damage Only (PDO) type accidents on overall results. Ranking of hot spots using
Gi* was done with the help of Z value associated with statistical significance. Ranking of hot spots using
K was done with the help of pixel values of various observed locations.

Results of hot spot analysis delineated various road stretches as well as intersections where hot spots
were concentrated. Results indicated that the estimation of hot spots by K and Gi* using three
conceptualization of spatial relationships (fixed distance band, inverse distance and inverse square
distance) are widely similar. Hot spots evaluated using different severity weighing systems were found to
be quite interesting. The results can be effectively utilized by various agencies for adopting better
planning and management strategies for accident reduction as well as for improved traffic operating
conditions.

Keywords: Accident hot spots, Getis-Ord Gi*, Kernel Density Estimation, spatial autocorrelation,
accident severity,

Introduction
Road accidents are considered one of the most negative impacts of developing modern
transportation system which result in injuries and loss of lives. Road traffic injury is the
eighth leading cause of death at the global level also it is leading cause of death for
young people aged 15-29 years [1]. India ranks at number 1 spot in the world in terms
of total road accident fatalities [1], which makes traffic safety as one of the most critical
matter in planning transportation strategy. Traffic safety engineers around the world
face one of the most imperative questions i.e., where to implement safety precautionary
measures so that they can have the most momentous impact for traffic safety. “Hot
spots”, “Black spots” or “High accident locations” are sites on the section of roads and
highways with higher accident frequency than expected at some threshold level of
significance. The most sensible solution to reduce accident frequency is accurate

ChoudharyAuthor1, OhriAuthor2 and KumarAuthor3

 3rd Conference of Transportation Research Group of India (3rd CTRG)

identification of hot spots. Hot spots can be identified using simple methods such as
Accident Frequency (AF) method, Accident Rate (AR) method, Accident Density (AD)
method and Accident Severity Index (ASI) method by various agencies around the
world. Some other methods for hot spot identification like Empirical Bayes (EB)
method [2], Public Participation Approach (PPA) [3], and Sequential pacing data
analysis technique [4] are also gaining popularity amongst researchers. EB method is
considered best hot spot identification method [5], but it requires special skills and
training in statistical analysis. The methods such as AF, AR, AD and ASI are less
superior in comparison to EB, but are simple and straightforward to use, that is why
many departments of transportation (DOTs) still rely on simple methods [6]. Recently
Geographic Information System (GIS) has made a noticeable impact in detecting road
accident hot spots. Spatial attribute combine with statistical analysis presents a superior
way to understand traffic accidents. GIS based techniques are relatively simple to use
and can convert raw statistical and geographical data into meaningful information for
spatial analysis, mapping and for identifying any factors contributing to accidents.
The primary objective of this paper is to compare two statistical techniques, Getis-Ord
Gi* [7] statistics and Kernel Density Estimation (K) using GIS for hot spot
identification. Previous studies done for identification of accident hot spots using
aforesaid techniques for Indian demographic and traffic conditions were solely based
on analysis using incidental points [8,9]. This work incorporated two different accident
severity weighing systems for analysis; this makes this work one of its own kind. Also,
three different Conceptualizations of Spatial Relationships (CSR) namely, fixed
distance band, inverse distance and inverse distance square were used to determine Gi*
and best CSR for hot spot identification. After comparing the results, the best method
and specified parameters for identifying hot spots were suggested generating optimum
results.It is envisaged that the findings of this study can be used to prioritize hot spots
and also help establish a correlation between surrounding environment and safety of
accident’s victims. This will help for successful traffic management and reduction of
traffic accidents.

Literature Review
GIS based accident information system can model relationships between spatial
phenomena that are nearly impossible to establish with non spatial database. Since
1900, GIS has been widely used by many researchers to geo-code accident locations,
developing pin maps of accidents and to perform database queries [10, 11]. Apart from
ranking accident locations as per higher accident rate, they can also be ranked
according to severity of accidents. Many researchers had assigned optimum weights to
accidents according to their severity, such as, Geurts et al. [12] proposed the weights of
5, 3 and 1 for fatal, grievous and minor accidents respectively. Similarly, RTA [13]
assigned weights of 3.0, 1.8, 1.3, and 1.0 for fatal, grievous, minor and property
damage only type accidents and Luathep [14] in his study proposed the weights of 125,
9 and 1 for fatal, injury and property damage only type accidents respectively.
Numerous methods have been developed for point pattern analysis and for detecting hot
spots. These methods can be classified under two categories [15]: a) methods which
analyze first-order effects, which calculate the variation in mean value of process such
as Kernel Density Estimation (K), quadrant count analysis etc; and b) methods which
examine second-order effects that calculates spatial autocorrelation of points for spatial
patterns, like Moran’s I Index, Geary C ratio and Getis-Ord Gi* statistics etc. Planer K
Choudhary, Ohri and Kumar
 3rd Conference of Transportation Research Group of India (3rd CTRG)

or simply K was utilized by researchers to identify hot spots for various roads and
highways [9, 16]. The major weakness of K is its inability to be tested for statistical
significance [17]. Spatial pattern of accident data could be analyzed analyzed by spatial
autocorrelation statistics which simultaneously combinescombine attribute similarity and
location proximity into single index. Unl
Unlike
ike K, statistical significance of these methods
can be evaluated with the Z scores. Getis-Ord Gi* statistic identifies the hot/cold spots
where features with significantly high/low values has to be surrounded by features
having simultaneously significant high/low values. Getis-Ord Gi* statistical method
was utilized to detect statistically significant traffic accident hot spots on various roads
and highways by researchers [8, 9, 18].

Study Area and Data Collection

Varanasi is situated on the banks of the mighty Ganges in the Indian state of Uttar
Pradesh, 320 kilometres south east of the state capitalcapita Lucknow. It is located at
25°16′55″ North Latitude and 82°57′23″ East Longitude. Mixed traffic composition and
narrow carriageways in old city area contribute to slow moving traffic. traffic. The main
accident prone areas are located on highways constituting fast moving vehicular traffic.
The study area selected is are area encompassing between NH-22 popularly known as
Kolkata-Delhi GT road, Varanasi
Varanasi-Kachhanwa SH-74, Varanasi- Sonauli NH NH-29 and
Varanasi- Kanyakumari NH-7. 7. Roads in these areas are always under extreme pressure
of traffic and have greater likelihood of accidents. Apart from these, areass falling under
jurisdiction of police stations namely Cantt, Ramnagar and Manduadih
Manduadih were also
included in analysis, where large numbers of accident prone or potential hot spot
locations were available. These areas contribute to large volume of traffic flow from
outer city areas to main city areas since these were located along the periphery o of the
main city. FIR’s in police records for five years from 2009 to 2013 were surveyed and
various police officers were interviewed to create accident database for analysis analysis.
Accident locations were plotted with the help of Google street maps and such locations
which were not properly specified in police records were cross checked by conducting
personal visits and by interviewing the local residents.

Figure:: 1 Study area and types of accident occurred in 2009

2009-2013
2013.

Choudhary, Ohri and Kumar

 3rd Conference of Transportation Research Group of India (3rd CTRG)

Methodology
Accident Datasets and Map Preparation
This study utilized ESRI’s ArcGIS 10.2 for analysis. The digital road map of Varanasi
was imported in Arc Map and saved as “roads” layer. Imported digital map and data
frame of Arc Map needed to have the same ‘projected co-ordinate’ system. The co-
ordinate system used for the present work is WGS 1980 UCS. The 5 years (2009-2013)
accident data of study area were collected from various police stations. Different
accident types such as fatal, grievous, minor and property damage only were reported
under clauses 304 (A), 338, 337 and 427 of the Indian Penal Code.
The accidents were geocoded by giving X and Y coordinates to each location. For
every geocoded point, a Feature Identity (FID) in a form of whole number is
automatically created on GIS. Every accident location was attributed with detailed
information such as type, landmark, month, date, time, vehicle type etc.

Severity Consideration
Studies were performed using accident incident points in order to determine high and
low clustering. But without weighted data, it is very difficult to determine whether the
observed clustering is true or not. To identify unsafe locations, crashes should be
weighted according to severity. In order to take severity of accident in consideration,
this study employs two different severity weighting systems alongside with incidental
accident data in analysis of high and low clustering.
Belgium system
This weighing system was adopted by Belgium government as a part of their official
methodology for hot spot detection. This system was successfully adopted to identify
black spot in three National Highways of Kerala in India [19]. Severity index for each
location can be calculated as per Equation 1:
SI1 = 5×X1 + 3×X2 + X3 (1)
where:
X1= total number of fatal accidents
X2= total number of grievous accidents
X3= total number of minor and property damage only accidents

New South Wales (NSW) system

In contrast of, Belgium system, this system provides discrimination between minor and
property damage type crashes [13]. It was successfully employed in identifying
pedestrian vehicle crash hot spot in streets of Adelaide in Australia [18]. The severity
index is calculated from following equation:
SI2 = 3.0×P+ 1.8×Q + 1.3×R+ S (2)
where:
P= total number of fatal accidents
Q= total number of grievous accidents
R= total number of minor accidents

Choudhary, Ohri and Kumar

 3rd Conference of Transportation Research Group of India (3rd CTRG)

S = total number of property damage only accidents

Kernel Density Estimation (K)

Kernel density is one of the important spatial analysis tools in commercially available
GIS software. K was utilized to calculate density of accidents within within a search
bandwidth of 0.3 km. K divides the entire study area into pre-determined
determined number of
cells. It uses a quadratic kernel function to fit a ssmoothly
moothly tapered surface to each
accident location as shown in Figure 2 [20]. The surface value reduces from the highest
at event location point to zero when it reaches radial distance from event location point.
The value of kernel function is assigned to every cell as individual cell values. The
resultant density of every cell is computed by adding its individual cell values. To
account accident severity,, the weight assigned
assign to each accident is represente
represented as its
Identification Number (ID). Population
P field of kernel density
ensity function is selected as
aforesaid ID. This facilitates count
counting of each accident according to its weight assigned.
In case of no severity, or analysis according to incident points, the population field is
selected as “None”. Kernel estimator can be defined as in Equation 3.

f(x) = ∑ ( ) (3)

where: h is termed as bandwidth, rad radius or smoothing factor; K is kernel and f is

estimator of probability density function. The kernel estimator depends upon choice of
bandwidth (h),, hence appropriate bandwidth should be determined according to
purpose of study.

Figure: 2 Principle of kernel density

density. Figure: 3 Spatial autocorrelation by function
[16] distance.

Spatial Autocorrelation:
utocorrelation: Moran’s I Method
M
Moran’s I is one of the oldest global spatial autocorrelation indicator which evaluates
whether the spatial pattern is clustered, random or dispersed. It works on both feature
locations and features values simultaneously. It combines the measure of locatiolocation
proximity and attributes similarity into an index. The index can be calculated using the
following Equation 4.
∑ ∑ ( ̅ )( ̅)
I=
I (4)
∑ ∑ (∑( ̅) )

where:
wij= the proximity weight of location i and j with wii = 0
Choudhary, Ohri and Kumar
 3rd Conference of Transportation Research Group of India (3rd CTRG)

̅ = the global mean value

xi= the severity index at location j
n = the total number of accident locations
The Incremental Spatial Autocorrelation (ISA) tool was used to calculate Moran’s I
index value and associated Z score, which represented statistical significance at
different threshold distances. The threshold distance with highest Z score was chosen
for mapping cluster using Getis-Ord Gi* function. The distance associated with highest
Z score gives the optimum clustering of high and low values. ISA tool gives the result
in the form of graphical representation between different distance threshold and their
associated Z score. Out of different values, either first peak or highest peak may be
chosen for optimum clustering. Figure 3 shows the different distance threshold with
their associated Z values. The choice of distances for analysis is solely based on spatial
distribution of accident locations. It is the essential condition for analysis that each
dataset should have at least one neighbour. The beginning distance for the analysis is
chosen as default value which is minimum value for which each accident location has
at least one neighbour. This distance was chosen to be 2100m with 50m increment
distance for analysis. The first peak was observed at 2150m as it is shown in Figure 3
which was subsequently chosen as distance threshold for optimum clustering in case of
fixed distance band CSR.

Getis-Ord Gi* Statistic

The two key processes involved in identification of desired hot spots were, collection
of events and mapping of clustering using Getis-Ord Gi* function. It creates a new
output feature class for every accident incident with a Z score and P value associated
with it. It identifies statistically significant hot spot as a location having high value and
surrounded by high valued neighbours as well. The local sum of values of a feature and
its neighbours is proportionally compared with sum of all features. When the local sum
obtained highly differs from expected local sum, and this difference is so high that it
couldn’t be a result of random chance, results in highly Z score and low P value. This
gives statistically significant high clustering.
In case of incident point data, weighted point feature class with field Icount was created
with help of collect event function. This indicated sum of all accidents happened in a
unique geographic location. This weighted point feature was used as input for running
Gi* function to identify whether features with high or low value tends to cluster in
study area. In case of analysis using severity weighing system, the weighted point
features also take severity of accidents in account. This was done by adding severities
of all accidents at a unique geographic location according to type of severity weighing
system. This weighted sum was used as input for running Gi* function. Getis-Ord Gi*
statistics and its standardized Z scores based on expected values E (Gi*) and the
variances (VAR (Gi*)) are mathematically expressed by equations (5) and (6)

∗ ∑ ( )
( )= ∑
(5)
∗
∗ − ( ∗)
= ( ∗)
(6)

where:
d = distance threshold
wij= weight of target neighbour pair
Choudhary, Ohri and Kumar
 3rd Conference of Transportation Research Group of India (3rd CTRG)

xj = severity index at location j

In case of fixed distance band CSR, distance threshold was chosen which associated
with high Z score. The distance associated with the first peak of graph between distance
threshold and Z score was chosen for mapping clusters. This is shown in figure (3). In
case of inverse distance and inverse squared distance CSR, the threshold distance was
set to “zero”, this indicated that all accident locations were considered neighbours to
each other.

Categorization of Hot spots

In case of analysis using Gi* statistics, all accident locations were categorized in four
categories based on their Z scores which signifies their statistical significance level.
The breaks were provided at Z scores of 1.65, 1.96 and 2.58, which shows statistical
significance level of 0.10,0.05 and 0.01 or confidence level of 90%, 95% and 99%
respectively. In case of analysis using K since there is no index associated with
statistical significance, hot spots were categorized employing Jenks algorithm. As
opposed to any arbitrary categorization scheme, this algorithm produces a group of
values that best represents the actual breaks observed in data, hence it conserves the
true clustering of data values [5]. In this study, the categorization was done in four
categories i.e., low, medium, high and very high priorities based on their associated
accident densities.

Results and its Discussions

No accidents were found in police records of old city area near Ganga’s ghats,
particularly near the old Kashi Vishwanath temple. This was due to absence of high
speed vehicles, predominance of pedestrian traffic and presence of police check posts
in the area. In other areas, police records revealed that the majority of accident
registered were either fatal (45.52 %) or grievous (40.75%). Obviously there was
negligence of people towards the filing of complaints for accidents of minor and
property damage only type which accounted for 8.15% and 5.56% respectively of total
accidents.
The results of hot spot identification done with the help of K and Gi* for different
accident severities and CSR’s are shown in Figures 4 to 6. They reveal locations of
potential hot spots within the study area of Varanasi city. Analysis was done with and
without considering the severity of accidents. Figure 4 shows the analysis without
severity consideration. It shows similar hot spots identified using K and Gi*, when
inverse distance and inverse distance square CSR was used. Severity was considered in
analysis using NSW and Belgium severity systems as shown in Figures 5 and 6. They
also show results similar to that without considering severity. This may be due to either
one of the reasons: a) due to contributing factors in accident such as speed of vehicle
and alignment of road; b) due to existence of any functional relationship between K and
Gi*. Since spatial location and severity of accidents were only parameters taken in
consideration, the plausible explanation is existence of functional relationship between
K and Gi*. Manepalli et al. [5] had earlier verified such functional relationship.
After analysis with and without severities using K and Gi* adopting inverse square and
inverse square CSR, 5 locations were identified as potential hot spots. These locations
are specified in Table 1 along with their respective ranking according to different

Choudhary, Ohri and Kumar

 3rd Conference of Transportation Research Group of India (3rd CTRG)

severities and CSR’s. These rankings were made according to Z score associated with
each location. Effect of accident severity over the results can be easily observed after
analyzing rankings obtained, with or without severity. Chawka ghat intersection has the
highest number of accidents and is the most significant hot spot when analyzed without
severity. But after considering severity according to NSW and Belgium systems,
Tengra intersection emerges as most significant hot spot. This is due to larger
percentage of fatal and grievous accidents.
Both severity weighing systems show identical rankings of hot spots. This is due to
lesser percentage of minor and PDO type accidents which do not cast their influence
over the final result. It is worthwhile to mention here that above results may have
undergone significant variation in case of appreciable percentage of minor and property
damage only crashes would have been recorded.
Results obtained from K and Gi* statistic are similar, but the latter produces results
associated with statistical significance. This makes Gi* statistic a better method than K
when analyzed using specific CSR’s. Inverse square and inverse distance square CSR
generated identical hot spots as shown in Figure 5 and 6 as well as in Table 1. Hot spots
identified from aforesaid CSR’s are distinctive as compared to those identified using
fixed distance CSR. This reveals that inverse distance and inverse distance square CSR
were better alternatives than fixed distance CSR.

Table 1: Rankings of various hot spots as per different severities and CSR.
Severity types Rankings of major hot spots based on GiZ score for various severities
& CSR
and CSR
Incident Points Belgium severity NSW severity
Hot Spots K Gi* Gi* Gi* K Gi* Gi* Gi* K Gi* Gi* Gi*
Identified (I) (I2) (F) (I) (I2) (F) (I) (I2) (F)
Tengra 4 2 2 X 1 1 1 X 1 1 1 X
Intersection
Chawkagha 1 1 1 X 2 2 2 X 2 2 2 X
t
Harsevanand 2 3 3 X 3 3 3 X 3 3 3 X
Intersection
Toll Plaza 5 4 4 X 5 4 4 X 5 4 4 X
Dafhi
Varuna 3 5 5 X 4 5 5 X 4 5 5 X
bridge
Note: Gi*(I) represents ranking due to Gi* statistic and Inverse distance CSR, Gi*(I2) represents ranking due to Gi*
statistic and Inverse distance square CSR, Gi*(F) represents ranking due to Gi* statistic and Fixed distance CSR and
X represents the insignificant ranking (Ranking not in range of 1 to 5).

Conclusions
The present work has exemplified use of GIS for processing of accident data and
performing complex spatial and statistical analysis using two different analysis
techniques K and Gi* for hot spot identification adopting various CSR’s. This study is
first of its kind in Indian traffic conditions taking severity of accidents in consideration.
Previous studies used aforesaid techniques for Indian traffic conditions, but were
limited to using single type of CSR without considering severity of accidents into
account. This study has revealed the superiority of Gi* statistic over K when used with

Choudhary, Ohri and Kumar

 3rd Conference of Transportation Research Group of India (3rd CTRG)

inverse distance and inverse distance square CSR’s for identification of hot spots
spots. This
method produced accurate and distinctive hot spots for given mixed traffic conditions.
Although this study is limited to identification of hot spots but it can also be utilized for
identification of cold spots. This analysis would have much easier,, accurate, descriptive
and reliable if accident records were more det
detailed
ailed and properly formatted. GPS should
be provided in every police station which facilitates recording of X and Y coordinates
of each accident location rather than name of nearest landmark. The accident severity
weighing systems used were adopted from co countries
untries having different traffic and
demographic conditions than India. Future research aims at development and utilization
of optimum severity weighing for hot spot identification.

Figure 4:: Hot spot identification using K and Gi* statistic for three CSR and
employing Incident points.
Choudhary, Ohri and Kumar
 3rd Conference of Transportation Research Group of India (3rd CTRG)

Figure 5:: Hot spot identification using K and Gi* statistic for three CSR and
employing NSW severity.

Choudhary, Ohri and Kumar

 3rd Conference of Transportation Research Group of India (3rd CTRG)

Figure 6:: Hot spot identification using K and Gi* statistic for three CSR and
employing Belgium severity.

References
1. World Health Organization,
Organization WHO global status report on road safety 2013:
supporting a decade of action.
action World Health Organization, (2013).
2. Hauer, E., Harwood, D.W., Council, F. M. and Griffith, M. S., Estimating
safety by the empirical Bayes method: a tutorial, Transportation Research
Record: Journal of the Transportation Research Board 1784, no. 1 (2002), pp.
126-131.
3. Kowtanapanich, W.,Tanaboriboon,
Tanaboriboon, Y. And Chadbunchchachai, W.,, Applying
process IATSS Research
public participation approach to hot spot identification process,
30 (1), (2005), pp. 73–
–85.

Choudhary, Ohri and Kumar

 3rd Conference of Transportation Research Group of India (3rd CTRG)

4. Mungnimit, S., Jierranaitanakit, K. and Chayanan, S., Sequential data analysis

for black spot identification, 4th IRTAD Conference, 2009,Seoul, pp. 219-222
5. Manepalli, U. R., Bham. G. H. and Kandada. S., Evaluation of hotspots
identification using Kernel Density Estimation (K) and Getis-Ord (Gi*) on I-
630, 3rd International Conference on Road Safety and
Simulation,2011,Indianapolis
6. Mitra, S., Spatial autocorrelation and Bayesian spatial statistical method for
analyzing fatal and injury crash prone intersections. Transportation Research
Record: Journal of the Transportation Research Board, No.2136, (2009), pp.
92-100.
7. Getis, A. and Ord, J. K., The analysis of spatial association by use of distance
statistics, Geographical analysis 24, no. 3 (1992), pp. 189-206
8. Prassanakumar, V., Vijith, H., Charutha, R. and Geetha, N., Spatio-temporal
clustering of road accidents: GIS based analysis and assessment, Procedia-
Social and Behavioral Sciences, 21, (2011), pp.317-325.
9. Rankavat, S. and Tiwari, G., Pedestrian Accident Analysis in Delhi using GIS,
Journal of the Eastern Asia Society for Transportation Studies 10, no. 0, (2013),
pp. 1446-1457.
10. Levine, N., Kim, K. And Nitz, L., Spatial Analysis of Honolulu Motor Vehicle
Crashes—Part I: Spatial Patterns, Accident Analysis and Prevention, Vol. 27,
No. 5, (1995), pp. 663–674.
11. Ghosh, S. K., Parida, M. and Uraon, J. K., Traffic accident analysis for
Dehradun city using GIS. ITPI Journal 1, No. 3, (2004), pp. 40-54.
12. Geurts, K., Wets, G., Brijs, T. and Vanhoof, K., Identification and ranking of
black spots: Sensitivity analysis, Transportation Research Record: Journal of
the Transportation Research Board 1897, no. 1, (2004), pp. 34-42.
13. RTA.Road Traffic Accidents in NSW - 1993. Sydney: Roads and Traffic
Authority of NSW,1994.
14. Luathep, P., Determination of economic losses due to road crashes in Thailand,
Master research study No-TE-03-2, Asian Institute of Technology, Bangkok,
2004.
15. O‘Sullivan, D., and Unwin, D. J., Geographic information analysis, John
Wiley, Hoboken, New Jersey, 2002.
16. Erdogan, S., Yilmaz, I., Baybura, T. and Gullu, M, Geographical information
systems aided traffic accident analysis system case study: city of
Afyonkarahisar, Accident Analysis & Prevention 40, no. 1, (2008), pp.174-181.
17. Anderson, T. K., Kernel density estimation and K-means clustering to profile
road accident hot spots, Accident Analysis and Prevention 41, no. 3, (2009), pp.
359-364.
18. Truong, L. T. and Somenahalli S. V. C, Using GIS to identify pedestrian-
vehicle crash hot spots and unsafe bus stops, Journal of Public
Transportation 14, no. 1, (2011), pp. 99-114.
19. Karuppanagounder, K., Black Spot Identification, Analysis and Improvement
Measures on Selected National Highway Stretches in Kerala, India, 90th Annual
Meeting of Transportation Research Board, (No. 11-1369), (2011).
20. Silverman, B. W., Density estimation for statistics and data analysis, Chapman
and Hall, New York, USA, 1986

Choudhary, Ohri and Kumar

View publication stats