Using data mining to improve traffic safety programs
| DOI | https://doi.org/10.1108/02635570610666412 |
| Pages | 621-643 |
| Published date | 01 June 2006 |
| Date | 01 June 2006 |
| Author | Scott Solomon,Hang Nguyen,Jay Liebowitz,William Agresti |
| Subject Matter | Economics,Information & knowledge management,Management science & operations |
Using data mining to improve
traffic safety programs
Scott Solomon, Hang Nguyen, Jay Liebowitz and William Agresti
Graduate Division of Business and Management, Department of Information
Technology, Johns Hopkins University, Rockville, Maryland, USA
Abstract
Purpose – The purpose of this paper is to demonstrate how the use of data mining (DM) analysis can
be used to evaluate how well cameras that monitor red-light-signal controlled intersections improve
traffic safety by reducing fatalities.
Design/methodology/approach – The paper demonstrates several different data modeling
techniques – decision trees, neural networks, market-basket analysis and K-means models.
Decision trees build rule sets that can abet future decision making. Neural networks try to predict
future outcomes by looking at the effects of historical inputs. Market-basket analysis shows the
strength of the relationships between variables. K-means models weigh the impact of homogenous
clusters on target variables. All of these models are demonstrated using real data gathered by the
Department of Transportation from fatal accidents at red-light-signal controlled intersections in
Maryland and Washington, DC from the year 2000 through 2003.
Findings – The results of the DM analysis will show predictable relationships between the
demographic data of drivers and fatal accidents; the type of collision and fatal accidents and between
the time of day and fatal accidents.
Research limitations/implications – The limitations of missing or incomplete data sets are
addressed in this paper.
Practical implications – This paper can act as a guide to follow for red light camera program
managers or local municipalities to conduct their own analysis.
Originality/value – This paper builds upon prior research in DM and also extends the body of
research that examines the effectiveness of red camera programs as they mature.
Keywords Decision trees,Neural nets, Statistical analysis, Databases, Road safety,
United States of America
Paper type Literature review
Effects of red light running
From 1992 to 2000, the number of fatal crashes at signalized intersections in the
United States increased by 19 percent (IIHS, 2001). Red light running (RLR) was
the single most frequent cause of these crashes, as pointed out by the Insurance
Institute for Highway Safety (IIHS, 2001) and equivalent to more than three times
the rate of increase for all other fatal crashes during the same period. According to
the Federal Highway Administration (FHWA), crash statistics show that nearly
1,000 Americans were killed and 176,000 were injured in 2003 due to RLR
related crashes. The monetary impact of crashes to society is approximately
$14 billion annually (FHWA, 2005). The California Highway Patrol estimates that
each RLR fatality costs the United States $2,600,000 and other RLR crashes cost
between $2,000 and $183,000, depending on severity (California State Auditor,
2002).
The current issue and full text archive of this journal is available at
www.emeraldinsight.com/0263-5577.htm
DM to improve
traffic safety
programs
621
Industrial Management & Data
Systems
Vol. 106 No. 5, 2006
pp. 621-643
qEmerald Group Publishing Limited
0263-5577
DOI 10.1108/02635570610666412
Installation of cameras
During the past 13 years there has been a surge in photo enforcement programs in
general and red light camera (RLC) programs in particular. The first RLC program in
the United States began in New York City in 1993 (FHWA, 2005). The New York City
program has become one of the largest RLC programs in the country, with more than
50 cameras (FHWA, 2005). Within a year of implementing its RLC program, New York
City issued 168,479 tickets with 15 cameras (New York City Red Light Camera
Program, 1997). To increase deterrence and spillover effects (spillover is when the
effects of cameras at intersections are felt at nearby intersections without cameras) the
city added an additional 200 fake cameras that flash but do not take actual pictures in
2002 (FHWA, 2002). The Institute of Transportation Engineers (1999) (ITE) reported
that the number of United States communities with operational automated red light
enforcement programs has increased from about 24 municipal enforcement pro grams
in 1998 to 30 communities in 1999 to 60 communities in 2001 (IIHS, 2001). According to
Blakely’s (2003) research public support for cameras reached 80 percent and the
number of communities where programs were in effect had expanded to 70.
Effect of cameras
Various studies have shown great improvements in traffic safety. Maccubbin et al.
(2001) prepared for the FHWA, in Oxnard, CA showed injury crashes at intersections
with traffic signals dropped 29 percent after camera enforcement began in 1997, Howard
county, Maryland reported that the number of crashes at every camera location were
reduced from 21 to 37.5 percent in the four years since camera operations began. A study
from Fairfax, Virginia, showed red-light violations declined 44 percent after one year of
camera enforcement; and likewise, Washington, DC red-light running fatalities
decreased from 16 to 2 percent in the first two years with red-light cameras; in Charlotte,
NC and New York City red-light running violations dropped by more than 70 and
62 percent in the first year, respectively; in San Francisco and Los Angeles, CA red-light
cameras led to a 68 and 92 percent reduction in the violation rate,, respectively, (FHWA,
2005). Time periods were not given for the two California studies.
Additionally, there was a spillover effect to nearby intersections not equipped with
cameras, indicating cameras will cause drivers to reduce RLR where the y think there
may be cameras, not just where they actually are (Blakely, 2003).
Effect of cameras internationally
Internationally, a five-year study in Australia showed fatal accidents dro pped
51 percent and injury-related accidents dropped 36 percent throughout 172 sites
monitored by 35 cameras (O’Connell, 2000). A three-year study in London with
25 cameras at 223 sites showed reductions of fatal accidents by 70 percent and
injury-related accidents by 28 percent (O’Connell, 2000). A study from Canada showe d:
...a 7% decline in crashes and up to 20% fewer deaths the first year the cameras were used.
The proportion of speeding vehicles at photo radar deployment stations in British Columbia
declined from 66% in 1996 to less than 40% today (Consumers Research Magazine, 1999).
In a Singapore study at three intersections, the average reduction in violation rates on
camera-monitored approaches was 40 percent, with a range of 4 to 63 percent (Lum and
Wong, 2003a, b).
IMDS
106,5
622
To continue reading
Request your trial