Using GIS Machine Learning Technique to Analysis Road Safety
near Parks
Yan Duan
Urban and Regional Planning Department, University at Buffalo-State University of New York, Buffalo, New York, U.S.A.
Keywords: Park Accessibility, Road Safety, Level of Traffic Stress, Machine Learning, Non-Motorized Transportation.
Abstract: Parks are essential components of the city fabric, providing recreation and relaxation places to residents. 0.25-
mile is a good walking distance. People within the park’s 0.25-mile service area are more likely to use non-
motorized transportation to travel to the park. In complex traffic settings, bicycles and pedestrians are
frequently at a higher risk of being involved in crashes. Thus, exploring road safety near parks is significant
for cyclists and pedestrians. This study aims to analyze and measure spatial relationships between traffic crash
clusters and park entrances to explore factors that influence traffic crashes near park entrances. The result
suggests that road traffic level of stress and roadway design and park parking locations are both factors to
impact the traffic conflicts near park entrances. To guarantee and improve safe non-motorized transportation
to parks, park planners and designers should consider alternative park accessibility and connectivity for
cyclists and walkers. This research aims to find factors that influence active transportation safety around parks.
The result of this paper will contribute to creating a safe walking and bicycle environment around parks.
1 INTRODUCTION
Parks provide a critical opportunity to strengthen the
fabric of a community, weaving together the social
and cultural landscape with the built and natural
environment. Research uncovered evidence of a
beneficial relationship between urban parks and
emotions, exercise, and attention (Kondo et al 2018).
As consequently, the COVID-19 epidemic has raised
the demand for active transportation, such as cycling
and walking, it provides an affordable, healthy, and
pollution-free daily option for transportation. In the
post-COVID Era, parks and public green nature
places have been gaining attractiveness as a
destination for active travel activities. Ensuring
accessibility and safety near parks and green spaces
is crucial. Also, the demand for active transportation
options like walking and cycling has grown within the
past ten years (Hasani et al 2019).
In urban places,
pedestrians and cyclists are frequently at a higher risk
of being involved in crashes due to complex traffic
situations. Plenty of research reveals that people
value safety beyond all else when choosing a means
of transportation. Although they love to ride, cyclists
feel that their commute is more dangerous and
vulnerable than that of automobiles, which is a major
impediment to riding and walking (Ferreira et al
2022). In urban settings, walking is a vital kind of
active transportation. Due to their vulnerability in
complex traffic conditions, people walking,
motorcycle riders, and pedal cyclists face a higher
chance of collisions in metropolitan areas than the
vehicle users.
Road safety is influenced by a multitude of factors,
including drivers, environmental conditions, and
vehicle specifications (Boggs et al 2020). Curb cuts
and junctions can have an impact on how frequently
and how badly cars collide with other cars, with
people, and with the environment. The type,
frequency, and severity of collisions are subsequently
impacted by these conflicts (Huang et al 2018).
Previous studies have explored park accessibility and
walkability through a built environment design
perspective, such as bike and pedestrian infrastructure,
parking space, transit stops, etc.
This study aims to explore park accessibility from
a traffic safety perspective. This paper examines
relationships between traffic crash location and park
entrance roadway level of traffic stress to explore
potential factors related to the crash locations near
parks. Traffic levels of stress can be a tool for planning
level analysis. The result can guide park designers to
think about design guidelines for park safety.
394
Duan, Y.
Using GIS Machine Learning Technique to Analysis Road Safety near Parks.
DOI: 10.5220/0012888000004547
Paper published under CC license (CC BY-NC-ND 4.0)
In Proceedings of the 1st International Conference on Data Science and Engineering (ICDSE 2024), pages 394-399
ISBN: 978-989-758-690-3
Proceedings Copyright © 2024 by SCITEPRESS Science and Technology Publications, Lda.
2 DATA AND STUDY AREA
2.1 Data Sources
Data used in our study areas are from multiple sources.
The table below illustrates the data set and sources.
Table 1 below explains the data and sources used in
this research. All data was downloaded as ArcGIS shp.
files. The crash data used in this study was from the
Georgia Department of Transportation between 2018
and 2022. A total of 152,517 accidents were collected
throughout the five-year range. The crash data includes
the location of all types of crashes, the roadway
alignment, surface conditions, weather conditions,
lighting conditions, and level of severity. Road
Centerline includes roadway speed limits, road names,
and road classification. The county boundary is the
exterior boundary of Gwinnett County. County Park
includes 53 developed parks in Gwinnett County.
Table 1: Research Data Source.
Data Source
Road
centerline
Atlanta Regional Commission Open Data
https://opendata.atlantaregional.com/
Traffic
Accident
Georgia Department of Transportation
https://gdot.aashtowaresafety.net/crash-data#/
County
boundary
Gwinnett County GIS Data Resource
https://gcgis-gwinnettcountyga.hub.arcgis.com/
County Park
Gwinnett County Online GIS Data
https://gcgis-gwinnettcountyga.hub.arcgis.com/
2.2 Study Area
The geographic scope of this research is within the
Gwinnett County Boundary. Fig. 1 shows the
Gwinnett County boundary, municipalities within the
County boundary, County Parks, major roads, and
non-residential land use.
3 METHODOLOGY
3.1 Data Process
This section presents the methodological component
of this study. A total of three methods are used in this
research. (1) GIS-mapping of the crash location by
using the hotspot cluster method. (2) Coding roadway
traffic level of stress based on the decision tree
method in GIS. (3) GIS-mapping road network for
visualizing park's 0.25 mi walking area.
3.1.1 GIS-Mapping Crash Hot Spot Analysis
(Getis-Ord 𝑮
𝒊
) (Spatial Statistics)
The Hot Spot Analysis is a tool for calculation of the
Getis-Ord 𝐺
statistic (pronounced G-i-star). The
resulting p-values and z-scores can show the spatial
clustering of characteristics with high or low values.
The Getis-Ord local statistic is given as:
𝑮
𝒊
=
,


,


,


,


(1)
Figure 1: Study Area-Gwinnett County (Picture credit: Original).
Using GIS Machine Learning Technique to Analysis Road Safety near Parks
395
Where 𝑥
is the attribute value for feather 𝑗, 𝑤
,
is the spatial weight between feature 𝑖 and 𝑗 , 𝑛 i s
equal to the total number of features, a n d 𝑋
a n d 𝑆
can be calculated through the following formulas (2)
and (3):
𝑿
=

(2)
𝑺=

𝑋
(3)
The 𝐺
statistic is a 𝓏 -socre so no further
calculations are required.
3.1.2 Level of Traffic Stress Method
An indicator named the Level of Traffic Stress (LTS)
differentiates a road network's features according to
how challenging it can be for walkers and bikers
(Huertas et al 2020).
When testing the perceived level of travel comfort,
Level of Traffic Stress is a useful measure for people
walking or bicycling along a given roadway.
Supporting efforts to develop safe and connected
networks of transportation facilities, LTS analysis can
identify streets that work well and areas in need of
improvement. Moving beyond minimum design
criteria, LTS helps planners, engineers, and advocates
understand the interrelated factors that either
encourage or discourage walking and bicycling.
Bicycle Level of Traffic Stress (BLTS) can be
used as a measurement for performance and safe with
respect to bicyclist. Pedestrian Level of Traffic Stress
(PLTS) is used to people who are neither riding on a
bicycle nor in a car. LTS is applicable regardless of
the presence or absence of a bike lane or sidewalk,
and it may be evaluated for both proposed and current
conditions.
The LTS's simplicity, which is based on a clear
decision tree approach in most of its implementations,
is one of its greatest advantages. However, because
there are so many segment-level variables required
for the categorization, LTS can be difficult to use and
comprehend (Harvey et al 2019). This research only
coded LTS for roadways near park entrances that have
a high traffic crash density cluster.
Table 2 below shows a matrix that determines
Bicycle Level of Traffic Stress and Pedestrian Level
of Traffic Stress for mixed traffic (no designated bike
lane, whether it has a shoulder or not). This table
shows a total of 4 levels of traffic stress, ranging from
1 to 4. There are 3 variables used for determining the
level of traffic stress: Traffic through lanes, annual
average daily traffic (AADT), and traffic speed (mph).
The degree of traffic stress is positively correlated
with all three variables. It suggests higher AADT,
greater velocity limits, and more lanes along roads
will all end up in higher levels of traffic stress.
3.2 Analysis of Roadway Network
Systems
Anyone may utilize network analysis to find service
areas surrounding a specific location on a network. A
network service area is a shed. A shed that contains
all accessible streets—that is, roadways that are
within a given impedance—is referred to as a network
service area. For instance, the 0.25-mile service area
for a park entrance includes all the streets that can be
reached within 0.25 miles (0.5 minutes walking) from
that park entrance. By doing this analysis, it can seen
that the 0.25-mile distance that people are most likely
to walk to parks. The average distance that an
American will walk before deciding to drive is 400
meters, or 0.25 miles or five minutes on foot. In US
park accessibility studies, a quarter mile (0.4 km) was
chosen as a typical threshold distance (Cutts et al
2009). In GIS, the study uses a network analysis tool
to create a 0.25-mile walking area for each park.
People who live or work within the 0.25-mile walking
area are more likely to use walking or bicycling to
access the park. These people have a higher risk of
being exposed to traffic crashes.
Table 2: Bicycle Level of Traffic Stress and Pedestrian Level of Traffic Stress Criteria.
Lanes AADT ≤ 20 mph 25 mph 30mph 35mph 40mph 45mph 50+ mph
1 thru lane
per direction
(or 1 lane
one-way
street)
0-750 1 1 3 4 4 4 4
751-1500 1 2 3 4 4 4 4
1501-3000 2 2 3 4 4 4 4
3000+ 2 3 3 4 4 4 4
2 thru lanes
per direction
0-7000 3 3 3 4 4 4 4
>7000 3 3 4 4 4 4 4
3+ thru lanes
p
er direction
Any ADT 4 4 4 4 4 4 4
ICDSE 2024 - International Conference on Data Science and Engineering
396
Figure 2: 2018-2022 Traffic Crash Point Map (Picture credit: Original).
Figure 3: 2018-2022 Traffic Crash Density Heat Map (Picture credit: Original).
Table 3: Parks have traffic crash clusters near their entrance.
Park Name Area (AC) Park Classification Street Name Bike
Lane
Ba
y
Creek Par
k
154.6 Communit
y
Par
k
Ozora R
d
No
Best Friend Par
k
43.1 Community Par
k
Jimmy Carter Blv
d
No
Club Drive Par
k
25.5 S
ecial Nei
hborhood Par
Club D
r
No
Duncan Creek Par
k
114.3 Community Par
k
Braselton Hwy Yes
Freeman's Mill Par
k
11.9 S
ecial Nei
hborhood Par
Alcov
y
R
d
No
Jones Bridge Par
k
30.0 Community Par
k
Jones Bridge R
d
No
Lilburn Activit
y
Buildin
g
2.1 Activit
y
Buildin
g
s Hillcrest R
d
No
Luck
y
Shoals Par
k
70.0 Communit
y
Par
k
Britt R
d
Yes
Mountain Park Par
k
43.4 Community Par
k
Five Forks Trickum R
d
No
Peachtree Rid
g
e Par
k
153.8 Communit
y
Par
k
Suwanee Creek R
d
No
Using GIS Machine Learning Technique to Analysis Road Safety near Parks
397
Pinckneyville Community
Cente
r
14.4 Community Recreation Centers Peachtree Industrial
Blv
d
Yes
Shorty Howell Par
k
66.5 Community Par
k
Pleasant Hill R
d
Yes
Sim
p
sonwood Par
k
222.4 O
p
en S
p
ace Par
k
Jones Brid
g
e Ci
r
No
South Gwinnett Par
k
22.9 Community Par
k
McGee R
d
No
Yellow River Post Office 5.1 Cultural Resource Park Five Forks Trickum Rd No
4 RESULT
4.1 County-Wide Level Mapping
Result Analysis
Fig. 2 shows county-wide crash locations. Fig. 3 is a
crash heatmap; it shows crash clusters. When some of
the crash cluster areas are within a park's 0.25-mile
walking shed, they bring safety risk to active
transportation.
4.2 Park Level Mapping Result
Analysis
This research finally selected 15 parks where traffic
crash clusters fall exactly on its main entrance. Then,
this paper selected the roadways that are next to park
entrances. The level of traffic stress results shows
most of the selected roadways are in traffic levels of
stress 4. This brings risk to pedestrians and cyclists
who travel to the 15 parks. In terms of the rest of the
county parks, there are traffic crash clusters within the
0.25-mile walking area of all county parks. Except for
those 15 parks that have higher traffic risk, all county
parks have potential traffic risk for cyclists and
walkers. Fig. 4 shows one example of the map result.
Figure 4: Bay Creek Park Entrance and Traffic Cluster Spot
(Picture credit: Original).
Table 3 shows detailed information on the fifteen
(15) parks that have high traffic risk for cyclists and
pedestrians. Most of them are community parks,
those park sizes are 25 acres or more. Community
Park typically is designed to serve an area within a ½
mile to over a 3-mile radius. Most of them do not
have separate infrastructure for active transportation.
In summary, based on the spatial analysis above,
there is a strong traffic crash impact near the park
entrance. And there is a correlation between the road
level of traffic stress and the density of crashes. A
higher and larger traffic crash density is on a higher
roadway level of traffic stress.
For deeper investigation, 226 traffic crashes that
occurred at the park entrance between 2018 and 2022
were gathered. Table 4 shows the crash detailed
information on the traffic crash near parks’ entrances.
It shows most of the crashes happened in clear
weather, daylight, and dry weather conditions. Most
of the injuries are not severe. The result may indicate
that under good weather conditions during the
daytime, people may be less careful when driving.
There is no data showing fatal crashes near park
entrances. There are, however, high levels of traffic
stress, such as high travel speed, multiple travel lanes,
and large traffic volumes that cause incidents when
turning into park entrance. To improve park
accessibility and safety for walkers and cyclers, there
should be more regulations around park entrances,
such as speed control, active and motorized
transportation entrance separation, control signals,
etc (Jerrett et al 2016 & Francis et al 2012).
Table 4: Detailed Information of 226 Traffic Crashes that
Near Park Entrance.
Surface Number
Dry 116
Wet 39
Water 3
Ice/Forst 1
Slush 1
Weather Number
Clear 138
Cloudy 53
Rain 33
ICDSE 2024 - International Conference on Data Science and Engineering
398
Snow 1
Sleet 1
Injury Number
No Injury 130
Possible Injury /
Complaint
80
Suspected Minor/Visible
Injury
13
Unknown 3
Light Number
Daylight 154
Dark-Lighted 31
Dark-Not-Lighted 21
Dawn 5
Dusk 5
5 CONCLUSION
This research uses GIS machine learning techniques
to explore traffic crashes and park accessibility and
safety for active transportation. Heatmap, street
network analysis, and roadway level of traffic stress
are three mythologies used for spatial analysis. This
study explores the connection between crash risk and
the quantity of traffic stress on the roads. A total of 53
developed county parks were used as research targets
and a total number of 152,517 traffic crashes
collected from 2018 to 2022 were used in this analysis.
The traffic heatmap shows a total of 15 park entrances
having traffic crash clusters. Roadways that are next
to those park entrances were analyzed in GIS for
traffic level of stress. The result shows that 12
roadways are in traffic stress level of 4, which is
categorized as the highest level of traffic stress. 2
roadways are in traffic stress level 3, which is
categorized as the second highest level. Simpson
Wood Park is an open space park where pedestrians
and bicycles rarely go, with few facilities and
amenities in the park, so people don’t consider that
park entrance roadway level. In planning park
entrances, a level of roadway traffic stress should be
considered to lessen the risk of crashes. Park entrance
should avoid being opened at roadways that are in
traffic stress level 4 or 3. Roadways that are in high
functional classification, with large number of
through lanes or having high speed limits will
negatively impact active transportation safety near
their park entrances, especially for parks that have
many facilities and amenities. A good number of
facilities and amenities will attract many visitors, and
this can increase traffic volume. Thus, to guarantee
park travel safety, within a park 0.25-mile walking
distance, regulations such as park zone speed control,
non-motorized and motorized entrance separation,
and traffic volume control should be considered when
designing a park. To guarantee that active travelers
receive the proper degree of care and access, laws
strategy, and planning efforts could be implemented
to address safe approaches to parks.
An additional aspect of the physical environment
that may promote a sense of community is public
areas, such parks, and plazas, which allow for chance
of interactions between neighbors. while offering
opportunities for people to access recreation and
nature. Walking and cycling to public places are not
only for improving residents’ physical and mental
health, but also saves energy to improve
environmental sustainability.
REFERENCES
M. Kondo, J. McKeon, C. Branas, International journal of
environmental research and public health, 15(3), 445
(2018).
M. Hasani, A. Jahangiri, IN. Sener, S. Munira, JM. Owens,
B. Appleyard, S. Ryan, SM. Turner, M. Ghanipoor,
Journal of advanced transportation, (2019).
M.C. Ferreira, P.D. Costa, D. Abrantes, J. Hora, S. Felicio,
M. Coimbra, TG. Dias, Transportation research part
F: traffic psychology and behaviour, 91:136-163
(2022).
A.M. Boggs, B. Wali, A.J. Khattak, Accident Analysis &
Prevention, 135:105354 (2020).
Y. Huang, X. Wang, D. Patton, Journal of transport
geography, 69:221-233 (2018).
J.A. Huertas, A. Palacio, M. Botero, GA. Carvajal, T. van
Laake, D. Higuera-Mendieta, SA. Cabrales, LA.
Guzman, OL. Sarmiento, AL. Medaglia,
Transportation Research Part D: Transport and
Environment, 85:102420 (2020).
C. Harvey, K. Fang, DA. Rodriguez, Mineta Transportation
Institute Publications (2019).
B.B. Cutts, K.J. Darby, C.G. Boone, A. Brewis, Social
science & medicine, 69(9):1314-1322 (2009).
M. Jerrett, J.G. Su, K.E. MacLeod, C. Hanning, D.
Houston, J. Wolch, Environmental research 151:742-
755 (2016).
J. Francis, B. Giles-Corti, L.Wood, M. Knuiman, Journal of
environmental psychology 32(4): 401-409 (2012).
Using GIS Machine Learning Technique to Analysis Road Safety near Parks
399