Identification of road traffic fatal crashes leading factors using principal components analysis

IJRET: International Journal of Research in Engineering and Technology eISSN: 2319-1163 | pISSN: 2321-7308
_______________________________________________________________________________________
Volume: 05 Issue: 01 | Jan-2016, Available @ http://www.ijret.org 79
IDENTIFICATION OF ROAD TRAFFIC FATAL CRASHES LEADING
FACTORS USING PRINCIPAL COMPONENTS ANALYSIS
Mohammad M Molla1
1
Graduate Research Assistant, Department of Civil and Environmental Engineering, North Dakota State University,
North Dakota, USA
Abstract
Traffic crash fatalities create primary safety concern beyond the traffic congestion and delay. Therefore, the purpose of this study
is to identify the principal components/factors associated with road traffic crash in the U.S. through retrospective reviewing based
on more than two million records of fatal crashes and 38 years (1975-2012) of National Highway Traffic Safety Administration
official’s Fatal Accident Reporting System (FARS) database. This study portrays an integrated geographic information system
and SAS application in order to find the major factors forcing traffic crashes. The resulting geospatial analysis and principal
components analysis yielded critical significant factors causing fatal traffic crashes. The outcomes of this research could be used
in transportation safety policy making and planning significantly.
Key Words: Accident Analysis Prevention, Clustering, Crash Hot Spot, Geographic Information Systems, Principal
Components Analysis, and Traffic Crash
--------------------------------------------------------------------***----------------------------------------------------------------------
1. INTRODUCTION
According to World Health Organization (WHO), by the
year 2030, the fifth most prevalent reason for deaths in the
world will be road traffic fatalities (Washington Post, 2013).
The United States has become third for traffic crash deaths
globally (WHO, 2013). The consequences of traffic crash
fatalities have a larger impact on global economy. In the
year 2012-13, the WHO, World Bank, U.S. Census Bureau,
Washington Post, and Forbes addressed the global economic
impact due to traffic crash fatalities and the economic
growth.
Traffic accident or crash factors are random and varied from
state to state. In order to understand the necessity of factors
associated with traffic crash and accident, a literature review
has been done. Traffic accidents and crime occurrence are
well defined threats to public safety (Kuo et al., 2013). Kue
et al. (2013) said that using data-driven procedures, police
departments would assigned constraint resources efficiently
in order to help crime and traffic crash safety, which
substantially reduce the crime and crashes in the hot spots
areas. U.S. DOT (2013) said “Some of life's greatest lessons
come when we learn from our mistakes, and in
transportation, where safety is of paramount importance,
that maxim is all the more true.” Despite the reality, U.S.
Department of Transportation and Federal Highway
Administration (FHWA) are strongly bound to reduce the
highway fatalities and injuries. According to FHWA (2012),
highway fatalities and injuries has been reduced
substantially from year 2007 to 2010 because of highway
safety programs influenced an important effects over the
nations. According to the traffic safety facts, it might be
inferred that after starting gathering traffic fatality crash data
since 1975, the traffic fatalities become declining for the
consecutive seven year (NHTSA, 2010).
States and municipalities must have accurate accident
modification factors so that maximum benefit of the capital
investment can be determined (NCHRP, 2005). According
to NCHRP (2005), “crash or accident reduction factors
(CRFs or ARFs) provide a quick way of estimating crash
reductions associated with highway safety improvements
and are used by many states and local jurisdictions in
program planning to decide whether to implement a specific
treatment and/or to quickly determine the costs and benefits
of selected alternatives”. Accident modification factors
(AMF) is facilitating AASHTO and NCHRP in order to
develop the strategic safety policy and guidelines so that
states and local users can be benefited. NCHRP also
discussed that the accident modification factors are always
obscured to the end user. In this regards, state agencies are
using their own developed system for depending on their
states crash data for the accident modification factors
planning. Thus the need of in-depth traffic crash/accident
analysis is an important issue is becoming more and more
important for the interest of federal, states, local agencies.
In this regards, since 1975 NHTSA maintaining the related
factors associated with fatal accident in to their structured
FARS databases. The FARS data element identifies factors
related to the crash expressed by the investigating officer
and associated in to three categories of related factors: 1)
Crash Related 2) Driver Related, and 3) Person Related.
The factors can be broadly categorized into several major
categories such as vision obscured; avoiding, swerving, or
sliding; roadway features; physical/mental condition;
distractions; non-motor-vehicle-operated related; and
miscellaneous causes.

_______________________________________________________________________________________
Traffic crash fatalities are geospatially related and enclosed
by clustered patterns for its geographic location. Statistics
revealed that the exponential increase of traffic fatalities are
concurrent with the economic growth in the nation’s history,
aggregated with widely spreading traffic accident hot spots.
The amount of traffic crash fatalities may eclipse the all-
time high set in the year 1976. Therefore, aggregated with
widely spreading traffic accident hot spots, the state
transportation agencies and law enforcement agencies are
working strategically towards implementing zero fatalities
on roadways.
Fig -1: Total Number of Fatalities per 100
Thousand People, 1980-2011
Figure 1 illustrates that the total number of traffic crash
fatalities per 100,000 people for states from 1980 to 2011.
Total number of fatalities was aggregated according to the
FARS published data. Based on 100,000 people, the highest
number of fatalities per population is in Wyoming and
Mississippi, which ranges from 753 to 904 fatalities per
100,000 populations in 38 years. The lowest number of
fatalities per 100,000 peoples includes the states of
Washington, New York, and New Jersey.
The fuzzy road network has multivariate factors which are
directly associated with traffic crash accident, which can be
overseen through the related factors of Fatal Accident
Reporting System (FARS) database from 1975-2012. FARS
database maintained different level of dataset such Vehicle,
Accident, and Person (Driver). In this study, our goal is to
identify the person (driver) related factors and detect any
possible clustering among the states. More specifically, the
primary goal of this research can be listed as to:
 Determine the leading person (driver) related
factors causing fatal traffic crash in the nation’s
highway.
 Detect any clustering among the U.S. States based
on the identified factors.
2. DATA AGGREGATION AND ANALYSIS
METHODOLOGY
The study area considered 48 states except District of
Colombia and Alaska. Accident data collected from
National Highway Traffic Safety Administration were
aggregated into the ArcGIS environment. Noted that the
data element identified factors related to the driver
expressed by an investigating officer. In the database,
related factors were stored in a different variable coded
varied with year. Thus, data was further processed and
cleaned up in a statistical software SAS. After cleaning up
the unnecessary data and clean up, final aggregated database
showed around two million crash events and 99 driver
related factors, which is presented in Table 1.
After processing the data, principal components analysis
was performed to identify the leading fatal crash factor and
based on the identified factors a single linkage cluster
analysis was performed in order to see if is there any
clustering among the 48 states. The cubic clustering
criterion (CCC), Pseudo F (PSF), and t2
(PST2) statistics are
used to determine the number of cluster for the data. SAS
(2008) discussed that the CCC and PSF are not suitable in
order to identify the number of clusters in a single linkage
method because the method has a tendency to shear the tails
of the distribution. In that case, PST2 could be utilized for
this purpose. Initially, any cluster with a large PST2 value
should be selected; the number of clusters used in the
analysis could be one greater than the initial number of
clusters.
Table -1: Driver Related Factors
SL Factors Description
00 None
01 Drowsy, Sleepy, Asleep, Fatigued
02 Ill, Passed Out/Blackout
03 Emotional (e.g., Depression, Angry, Disturbed)
04 Reaction to or Failure to Take Drugs/Medication
05 Under the Influence of Alcohol, Drugs, or Medication
06 Inattentive/Careless (Talking, Eating, Car Phones,
etc.)
07 Restricted to Wheelchair
08 Road Rage/Aggressive Driving
09 Impaired Due to Previous Injury
10 Deaf
11 Other Physical Impairment
12 Mother of Dead Fetus/Mother of Infant Born Post
Crash
13 Mentally Challenged
14 Failure to Take Drugs/Medication
15 Seat Back Not in Normal Position, Seat Back Reclined
16 Police or Law Enforcement Officer
17 Running off Road
18 Traveling on Prohibited Traffic ways
19 Legally Driving on Suspended or Revoked License
20 Leaving Vehicle Unattended with Engine Running;
Leaving Vehicle Unattended in Roadway
21 Overloading or Improper Loading of Vehicle with
Passenger or Cargo

_______________________________________________________________________________________
22 Towing or Pushing Vehicle Improperly
23 Failing to Dim Lights or to Have Lights on When
Required
24 Operating Without Required Equipment
25 Creating Unlawful Noise or Using Equipment
Prohibited by Law
26 Following Improperly
27 Improper or Erratic Lane Changing
28 Failure to Keep in Proper Lane
29 Illegal Driving on Road Shoulder, in Ditch, or
Sidewalk, or on Median
30 Making Improper Entry to or Exit from Traffic way
31 Starting or Backing Improperly
32 Opening Vehicle Closure into Moving Traffic or
Vehicle is in Motion
33 Passing Where Prohibited by Posted Signs, Pavement
Markings, Hill or Curve, or School Bus Displaying
Warning Not to Pass
34 Passing on Wrong Side
35 Passing with Insufficient Distance or Inadequate
Visibility or Failing to Yield to Overtaking Vehicle
36 Operating the Vehicle in an Erratic, Reckless, Careless
or Negligent Manner or Operating at Erratic or Suddenly
Changing Speeds
37 Police Pursuing this Driver or Police Officer in Pursuit
38 Failure to Yield Right of Way
39 Failure to Obey Actual Traffic Signs, Traffic Control
Devices or Traffic Officers, Failure to Observe Safety
Zone Traffic Laws
40 Passing Through or Around Barrier
41 Failure to Observe Warnings or Instructions on
Vehicle Displaying Them
42 Failure to Signal Intentions
43 Driving too Fast for Conditions
44 Driving Too Fast for Conditions or in Excess of
Posted Speed Limit
45 Driving Less Than Posted Maximum
46 Operating at Erratic or Suddenly Changing Speeds
47 Making Right Turn from Left-Turn Lane or Making
Left Turn from Right-Turn Lane
48 Making Improper Turn
49 Failure to Comply With Physical Restrictions of
License
50 Driving Wrong Way on One-Way Traffic way
51 Driving on Wrong Side of Road (Intentionally or
Unintentionally)
52 Operator Inexperience
53 Unfamiliar With Roadway
54 Stopping in Roadway (Vehicle Not Abandoned)
55 Under riding a Parked Truck
56 Improper Tire Pressure
57 Locked Wheel
58 Over Correcting
59 Getting Off/Out of or On/In to Moving Vehicle
60 Getting Off/Out of or On/In to Non-Moving Vehicle
61 Rain, Snow, Fog, Smoke, Sand, Dust
62 Reflected Glare, Bright Sunlight, Headlights
73 Driver Has Not Complied with Learners Permit or
Intermediate Driver License Restrictions (GDL
Restrictions, Since 2004)
74 Driver Has Not Complied With Physical or Other
Imposed Restrictions (Since 2004)
75 Broken or Improperly Cleaned Windshield
76 Other Obstruction
77 Severe Crosswind
78 Wind from Passing Truck
79 Slippery or Loose Surface
80 Tire Blow-Out or Flat
81 Debris or Objects in Road
82 Ruts, Holes, Bumps in Road
83 Live Animals in Road
84 Vehicle in Road
85 Phantom Vehicle
86 Pedestrian, Pedal cyclist, or Other Non-Motorist in
Road
87 Ice, Water, Snow, Slush, Sand, Dirt, Oil, Wet Leaves
on Road
88 Trailer Fishtailing or Swaying
89 Carrying Hazardous Cargo Improperly
90 Hit-and-Run Vehicle Driver
91 Non-Traffic Violation Charged – Manslaughter or
Homicide or Other Assault
92 Other Non-Moving Traffic Violation
93 Cellular Telephone
94 Cellular Telephone in Use in Vehicle
95 Computer Fax Machines/Printers
96 On-Board Navigation System
97 Two-Way Radio
98 Head-Up Display
99 Unknown
3. RESULTS AND DISCUSSIONS
3.1. Principal Components Analysis
The Principal Components Analysis was completed using
the statistical SAS programming. The first eigenvalue of
236.995, accounts for approximately 39.57% of the
standardized variance. A total of 99 factors were used
during PCA analysis. It was revealed that only thirteen
eigenvalues explained the 80% of the variance, which is
presented in Table 2 and Figure 2. PCA analysis suggests
that the first thirteen principal components provide an
adequate summary of the data for most purposes. This
thirteen principal components discovered for 45 factors,
which is substantially influence on the traffic crash
accidents.
Table -2: Eigenvalues of Principal Components Analysis
Eigenvalues of the correlation matrix: total=599, average=1
SL Eigenvalue Difference Proportion Cumulative
1 236.995207 186.713767 0.3957 0.3957
2 50.281440 15.719126 0.0839 0.4796
3 34.562314 9.715977 0.0577 0.5373
4 24.846337 3.905710 0.0415 0.5788
5 20.940627 1.910206 0.0350 0.6138
6 19.030421 1.445640 0.0318 0.6456
7 17.584781 2.975183 0.0294 0.6750

_______________________________________________________________________________________
8 14.609598 1.740300 0.0244 0.6994
9 12.869298 0.028081 0.0215 0.7209
10 12.841217 0.790838 0.0214 0.7423
11 12.050379 0.250035 0.0201 0.7624
12 11.800344 0.911742 0.0197 0.7821
13 10.888602 1.248457 0.0182 0.8003
Fig -2: Scree Plot of Factors Selection
Interestingly in the first principal component, out of 45
factors, only 23 factors contributing significantly which is
presented in the Table 3. Result showed that high speed is
the number one cause for fatal traffic accident in the 48
states. High speed, Starting and backing improperly, failure
to observe warnings or instructions, passing with
insufficient distance, and driving less than posted speed
limit is the top five fatal traffic accident factor. Result also
showed that there are other factors identified in the top 23
factors which is related to unconsciousness of driver such as
depression and inattentive.
Table -3: Top Influential Factors
Factors Positi
on
High-Speed Chase with Police in Pursuit 1
Starting and Backing Improperly 2
Failure to observe Warnings or Instructions 3
Passing with insufficient distance 4
Driving less than posted speed limit 5
Towing and pushing vehicle improperly 6
Operating the vehicle in an erratic, reckless,
careless and negligent manner
7
Creating unlawful noise 8
Failure to Take Drugs/Medication 9
Mother of Dead Fetus 10
Running off Road 11
Other Drugs(Marijuana, Cocaine) 12
Deaf 13
Operating without required equipment 14
Depression 15
Passing on wrong side 16
Seat Back Not in Normal Position 17
Mentally Challenged 18
Leaving vehicle unattended 19
passing where prohibited by posted signs 20
Inattentive 21
Driving in excess of posted speed limit 22
Illegal driving in road shoulder, ditch and sidewalk
or median
23
3.2. Selection of Number of Clusters
Using single linkage cluster procedure, hierarchical cluster
of observation was initially determined. Using the top 45
factors which may support to explain the 80% of the
variance of the data was selected in order to select the
number of cluster. Table 4 provides the results portraying
the last 30 generations of the cluster history for the single
linkage method.
Table -4: Cluster History
Number
of
Clusters
Cubic
Clustering
Criterion
Pseudo F
Statistic
Pseudo
t-Squared
Norm
Minimum
Distance
Tie
30 . 29.4 3.0 0.0556 T
29 . 26.1 3.0 0.0556 T
28 . 23.4 3.0 0.0556 T
27 . 21.2 3.0 0.0556 T
26 . 19.3 3.0 0.0556 T
25 . 17.8 3.0 0.0556 T
24 . 16.4 3.0 0.0556 T
23 . 15.3 3.0 0.0556 T
22 . 14.4 3.0 0.0556 T
21 . 13.5 3.0 0.0556 T
20 . 12.8 3.0 0.0556 T
19 . 12.2 3.0 0.0556 T
18 . 11.7 3.0 0.0556 T
17 . 11.3 3.0 0.0556 T
16 . 12.4 . 0.0556 T
15 . 13.7 3.0 0.0556 T
14 . 15.2 3.0 0.0556 T
13 . 16.9 3.0 0.0556 T
12 . 18.8 3.0 0.0556 T
11 . 21.1 3.0 0.0556 T
10 . 23.9 3.0 0.0556 T
9 -15 27.6 . 0.0556 T
8 -13 32.3 3.0 0.0556 T
7 -12 38.5 3.0 0.0556
6 -10 46.6 24.0 0.1112 T
5 -8.2 57.8 9.6 0.1112 T
4 -9.8 27.8 41.5 0.1112 T
3 -8.8 23.4 15.8 0.1112
2 -4.9 47.1 4.6 0.1668 T
1 0.00 . 47.1 0.1668
Figure 3 plots all three statistics based on the number of
clusters. From Table 4 and Figure 3, it is evident that peak
values of PST2 found at level 4 and 6. Thus the number of
cluster will be 4+1=5, or 6+1=7. But in the last column of
Table 4, we can see that there is ties exist for cluster 5,
which means at this level, estimation of number of clusters
would be indeterminate. Therefore, at level 6 the number of
clusters might be suitable as 7. Finally, seven clusters were
selected based on the 45 factors. The results of the
hierarchical clustering procedure have been presented
visually using a tree diagram/dendogram in Figure 4. The

_______________________________________________________________________________________
dendogram displays all the stages of the hierarchical
procedure and the distances at which clusters were merged.
Dendogram also suggest for only seven clusters that be
considered. The computed clustered states have been
presented in choropleth map in Figure 5. It was also
revealed that Texas, California, Mississippi, Florida,
Pennsylvania, and Ohio has extreme scenario in case of
traffic fatalities.
Fig -3: Criteria for the Number of Clusters
Fig -4: Dendogram Showing Minimum Distance between
Clusters
Fig -5: Clusters of States
4. CONCLUSIONS
It might be inferred that based on 38 years of FARS data
only 23 drivers related factors contributing significantly or
has substantial influence on the nationwide traffic crash
accidents/incidents, this has been listed in the Table 3. Out
of these top 23 leading factors, top ten factors are high-
speed chase with police in pursuit; starting and backing
improperly; failure to observe warnings or instructions;
passing with insufficient distance; driving less that posted
speed limit; towing and pushing vehicle improperly;
operating the vehicle in an erratic, reckless; careless and
negligent manner; creating unlawful noise; failure to take
drugs/medication; mother of dead fetus. Interestingly high-
speed chase with police in pursuit; driving less than posted
speed limit; driving in excess of posted speed limit has been
revealed in the top 23 leading driver related factors causing
traffic accident in the United States. Study also discovered
that Drugs related to marijuana, cocaine, etc. are the twelfth
most reason for the traffic accident in the United States.
Inattentive while driving is 21th reason for traffic accident.
The main objective of this research was to see if is there any
clustering relationship exist in 48 states considering the
driver related most important 45 factors. The computed
clusters based on single linkage method revealed that
considering 45 driver related factors which causing accident,
may classify the 48 states into seven clusters. Number of
clusters found technically acceptable but not providing
enough information about the remaining 42 states. Since
Texas, California, Mississippi, Florida, Pennsylvania, and
Ohio has large number of traffic fatalities, this six states are
forming six different clusters. These six states are
influencing the clustering analysis. Remaining 42 states may
be analyzed further to see how the cluster forms without the
effect of the larger six states. Due to resources constraints,
the goal of the study was limited as specified earlier;
however, many cause and effect relationships can be
established based on these 38 years data. Thus, further
research can be performed considering combining the
driver, vehicle and crash related factors.

_______________________________________________________________________________________
REFERENCES
[1] FHWA (Federal Highway Administration-U.S.
Department of Transportation). (2012).
<http://safety.fhwa.dot.gov/provencountermeasure
/pc_memo.htm> (Accessed December, 2013).
[2] Kuo, P., Lord, D. and Walden, T. D. (2013). “Using
geographical information systems to organize police
patrol routes effectively by grouping hotspots of crash
and crime data.” Journal of Transport Geography, 30,
138-148.
[3] Molla, M. M., Stone, M. L., Lee, E. (2014).
“Geostatistical Approach to Detect Traffic Accident
Hot Spots and Clusters in North Dakota”. Upper Great
Plains Transportation Institute.
[4] NCHRP (National Cooperative Highway Research
Program). (2005). “Crash Reduction Factors For Traffic
Engineering and Intelligent Transportation System (Its)
Improvements: State-Of-Knowledge Report.” Research
Results Digest 299.
[5] NHTSA (National Highway Traffic Safety
Administration). (2010).
<http://wwwnrd.nhtsa.dot.gov/Pubs/811363.pdf>
(accessed December, 2013).
[6] Rencher, A. C. (2001). “Methods of Multivariate
Analysis.” Second Edition, John Wiley & Sons, Inc.,
New York, NY. SAS (SAS Institute Inc.). (2008).
“SAS/STAT® 9.2 User’s Guide.” Cary, NY.
[7] USDOT (United States Department of Transportation
Research and Innovative Technology Administration).
(2013).
<http://www.rita.dot.gov/sites/default/files/rita_archive
s/rita_publications/horizons/2008_11_12/html/improvin
g_operational_safety.html> (Accessed December,
2013).
[8] Washington Post. (2013).
<http://www.washingtonpost.com/blogs/worldviews/wp
/2013/01/18/a-surprising-map-of-countries-that-have-
the-most-traffic-deaths/> (Accessed December, 2013).
[9] WHO (World Health Organization). (2013).
<http://www.who.int/violence_injury_prevention/road_
traffic/en> (Accessed December, 2013).

Identification of road traffic fatal crashes leading factors using principal components analysis

More Related Content

Similar to Identification of road traffic fatal crashes leading factors using principal components analysis (20)

More from eSAT Journals (20)

Recently uploaded (20)

Identification of road traffic fatal crashes leading factors using principal components analysis