Analytical Techniques for Evaluating the Implementation

of Adaptive Traffic Signal Control Systems

Abhay D. Lidbe, Ph.D. 1; Elsa G. Tedla 2; Alexander M. Hainen, Ph.D. 3;
and Steven L. Jones Jr., Ph.D., M.ASCE 4
Downloaded from ascelibrary.org by Universidad Nacional Autonoma de Mexico on 02/25/19. Copyright ASCE. For personal use only; all rights reserved.

Abstract: Adaptive traffic signal control (ATCS) reduces congestion, improves travel time reliability, and prolongs the effectiveness of
traffic signal timing. The objective of this paper is to demonstrate different data sources and analysis techniques to compare the arterial
performance for before, during, and after ATCS implementation. This paper presents several analyses using field-collected travel time data
and simulation modeling techniques. The paper demonstrates that different sources of travel time data can be used to characterize the corridor
performance under ATCS. Although corridor travel time is the traditional way to assess the improvements from the newer technologies, other
measures need to be looked at for objective evaluation of the new systems. It is recommended to use multiple evaluation techniques for better
decision making as every technique has its own limitations. The methodologies discussed will help transportation engineers, planners, and
policymakers to understand what analysis can be conducted at any point of an ATCS deployment to evaluate the system performance.
As policies and timings are adjusted, this feedback is an effective way to continuously monitor the ATCS and quantify benefits. DOI:
10.1061/JTEPBS.0000034. © 2017 American Society of Civil Engineers.
Author keywords: Adaptive traffic signal control (ATCS); Crowdsourced data; Bluetooth data; Microsimulation.

Introduction and Background
According to the Federal Highway Administration, traffic conges-
tion costs Americans approximately 6 billion hours, and poor signal
timing contributes about 5% of total congestion (Grant et al. 2010).
The National Traffic Signal Report Card emphasizes a greater need
for better signal management and operations (National Transporta-
tion Operations Coalition 2012). Traffic signal operation is a criti-
cal component of efficient traffic management. Proper traffic signal
timing and operations reduce congestion, improve mobility, and
enhance safety. The 2015 Urban Mobility Scorecard suggests sim-
ple actions, such as retiming of traffic signals, as a means for con-
gestion relief (Schrank et al. 2015). Signal retiming is often
effective for a while and will eventually require subsequent reviews
and periodic updating. This process of periodic retiming can be
avoided by using adaptive traffic control systems (ATCS).
ATCSs continuously detect the vehicular traffic on the road-
ways. These systems then adjust the signal timing plan in real time
to adapt to the changing traffic conditions. ATCSs are believed to
benefit the users by reducing delay, congestion, and fuel consumption.
1 strategies provide similar performance. However, during the shoulder
Dept. of Civil Construction and Environmental Engineering, Univ. of
Alabama, P.O. Box 870205, Tuscaloosa, AL 35487-0205 (corresponding
author). E-mail: adlidbe@crimson.ua.edu
2
Research Engineer, Dept. of Civil Construction and Environmental
Engineering, Univ. of Alabama, P.O. Box 870205, Tuscaloosa, AL
35487-0205. E-mail: egtedla@eng.ua.edu
3
Assistant Professor, Dept. of Civil Construction and Environmental
Engineering, Univ. of Alabama, P.O. Box 870205, Tuscaloosa, AL
35487-0205. E-mail: ahainen@eng.ua.edu
4
Associate Professor, Dept. of Civil Construction and Environmental
Engineering, Univ. of Alabama, P.O. Box 870205, Tuscaloosa, AL
35487-0205. E-mail: sjones@eng.ua.edu
Note. This manuscript was submitted on June 8, 2016; approved on
October 26, 2016; published online on February 14, 2017. Discussion
period open until July 14, 2017; separate discussions must be submitted
for individual papers. This paper is part of the Journal of Transportation
Engineering, Part A: Systems, © ASCE, ISSN 2473-2907.
They also have the potential to improve travel time reliability.
The 2012 Urban Mobility Report published by Texas A&M Trans-
portation Institute reports ATCS systems as performing some three
times better than actuated control with regard to delay reduction
(Schrank et al. 2012). Previous evaluations of ATCS implementa-
tion at different locations have corroborated such benefits (Martin
and Stevanovic 2008; Shelby and Bullock 2008; Tian et al. 2011).
However, according to one such study, gains in travel time were
achieved by ATCS over an arterial corridor (Hutton et al. 2010),
but the improvements were directionally specific and were also
time specific. Another study shows no significant changes in the
travel times before and after ATCS deployment under certain con-
ditions (SRF Consulting Group, Inc. 2000). Improvement in arterial
performance at the expense of side-street efficiency was found in an
evaluation of ATCS deployment in Florida (Gord & Associates
2007). Another study simulated the semiactuated, coordinated,
time-of-day (TOD) using hardware-in-the-loop simulation (HILS)
and the Sydney Co-ordinated Adaptive Traffic Control System
(SCATS) on an 11-intersection arterial of Cobb Parkway in Cobb
County, Georgia (Hunter et al. 2010). The initial findings from the
comparative results indicate that during peak conditions, both control
peak periods, ATCS provides better traffic control suitable to the cur-
rent conditions. In one such before-after ATCS comparative study, the
authors are of the view that a biased TOD plan can influence the
ATCS performance results (Peters et al. 2007). The literature review
shows mixed results for ATCS deployments. Thus, ATCS perfor-
mance needs to be appropriately evaluated at any point in time.
Deployments of newer technologies, such as ATCSs, are often
expensive, complex, and require a lengthy implementation period
that can make it difficult for engineers to assess improvements over
time. ATCS installations can take about 18 months (Stevanovic 2010)
or longer. Failure to reconfigure the ATCS systems for the possible
changes in traffic conditions, geometrical features, or transportation
policies during this period can lead to loss of ATCS performance.
Additionally, it is often misperceived that once installed, ATCS sys-
tems can be left to operate with little maintenance (Stevanovic 2010).

© ASCE 04017011-1 J. Transp. Eng., Part A: Syst.

J. Transp. Eng., Part A: Systems, 2017, 143(5): 04017011

 An ATCS may underperform if it is not customized or not fine-tuned
to specific policy requirements. Stevanovic (2010) also notes that
there is a need for comprehensive evaluation study to understand the
true costs and benefits of ATCSs both during and post-deployment.
Such evaluations and continuous monitoring help transportation plan-
ners to better understand the effectiveness of their investment strat-
egies, fine-tune the operations of implemented projects, and calibrate
or refine their planning tools and models (Grant et al. 2010). Further,
it is recommended multiple data sources be considered depending on
the analysis performed and the level of detail required (Tahmasseby
2015). Considering these aspects, this paper compares the arterial
Downloaded from ascelibrary.org by Universidad Nacional Autonoma de Mexico on 02/25/19. Copyright ASCE. For personal use only; all rights reserved.
department. After the project has been selected, planning and de-
signing the ATCS begins in the design phase (ϕii). Design traffic
observations and preliminary performance measures can be col-
lected and simulation of the ATCS can be conducted for these be-
fore conditions. After the design is finalized, the ATCS hardware
and software will be installed and detailed before observations
should be collected (ϕiii). After all the components are in place,
the system can be turned on and short-term comparisons can be
conducted for preliminary assessment and adjustment (ϕiv). A
burn-in period for the hardware and software (making sure that

all the equipment is operating, communicating, and functioning

performance and demonstrates different analytical techniques and
methods to evaluate ATCS performance at different stages of its
implementation using data from different sources.
Fig. 1 shows a typical timeline for implementation of an ATCS.
While it is important to monitor ATCS performance at any point
during the implementation process to fully recognize its impor-
tance, it is equally important to assess ATCS benefits before, dur-
ing, and after implementation. Typically, ATCS will be identified
as a suitable solution for a corridor improvement project at the
project programming level (ϕi). This is often based on preliminary
volumetric data and other information from the traffic operations
as expected) will follow to examine weekdays, weekends, different
time-of-day settings, and other hardware and software configura-
tion groups (ϕv). Once the ATCS is fine-tuned and fully functional,
additional policy changes, tunings, and adjustments are made by
the local traffic engineering department as warranted to manage
the future traffic changes (ϕvi). Finally, long-term monitoring
can be conducted (ϕvii). It should be noted that ϕvi and ϕvii
are iterative in nature since any changes or adjustments made to
the ATCS will require monitoring the changes in its performance.
Throughout all of these periods, Fig. 1 discusses different analyses
that may be performed.

Fig. 1. Adaptive traffic signal control implementation process and practices

© ASCE 04017011-2 J. Transp. Eng., Part A: Syst.

J. Transp. Eng., Part A: Systems, 2017, 143(5): 04017011

Downloaded from ascelibrary.org by Universidad Nacional Autonoma de Mexico on 02/25/19. Copyright ASCE. For personal use only; all rights reserved.
Fig. 2. (Color) Summary of SCATS deployment case study
Purpose and Scope
The purpose of this paper is to demonstrate the use of different
data sources and analytical techniques to compare the arterial
performance for before, during, and after ATCS implementation
using field-collected travel time data and simulation modeling. Spe-
cifically, the paper illustrates the use of different techniques to
evaluate an ATCS after deployment (from turn-on through long-
term monitoring in Fig. 1) based on a case study in Montgomery,
Alabama. The approach discussed will help transportation engi-
neers, planners, and policymakers to understand various analyses
that can be conducted at different stages of an ATCS deployment to
evaluate the system performance. As policies and timings are ad-
justed, this feedback is an effective way to continuously monitor
the ATCS.
Case Study
The case study corridor shown in Fig. 2 is US-231/Eastern
Boulevard in Montgomery, Alabama. The corridor is a 3.7 km
(2.3 mi)–long six-lane arterial with 10 signalized intersections.
The average spacing between adjacent intersections is 0.42 km
(0.26 mi). The corridor experiences average traffic volumes of
1,900 vehicles per hour (vph) per intersection on the main line
and approximately 150 vph on cross streets during typical weekday
PM peak hour (16:30–17:30). Based on the field observations and
traffic conditions, the saturation level was subjectively assessed to
be moderate. Three time periods—AM peak (7:30–8:30), PM peak
© ASCE 04017011-3
J. Transp. Eng., Part A: Systems, 2017, 143(5): 04017011
(16:30–17:30), and off-peak (13:30–14:30) are considered for the
study’s purpose.
The ATCS system evaluated in the case study is SCATS. This
paper presents an evaluation of the SCATS deployment after it was
turned on in October 2012. A timeline of the SCATS deployment
and a schematic map of the corridor are shown in Fig. 2.
Data Requirements
Data for ATCS implementation (or other corridor signal projects)
can vary from turning movement counts, link volumes, floating car
travel times, and many other different sources. The types of studies
depend on the policies and goals that are examined before-ATCS
and after-ATCS installation. Typically, travel time, delay, travel
speed, queue lengths, number of stops, and level of service are used
to evaluate the performance of movements along the corridor.
Collecting travel time data is one of the most critical pieces of
data used for assessing the main line corridor performance. The
accuracy of travel time data also has high implications since even
the smallest inaccuracies can sway the research findings or the
planner’s decision. The proliferation of wireless technologies
and mobile devices has opened up the entire space for using
and collecting travel time data. Along with the traditional field data,
two other major prevalent state-of-the-art data collection techniques
are (1) Bluetooth media access control (MAC)-address matching;
and (2) commercially available crowdsourced data. Table 1 sum-
marizes different data collected using different data sources, data
collection periods, and their usage in this study.
J. Transp. Eng., Part A: Syst.
 Table. 1. Summary of the Study-Specific Data and Its Sources, Collection Period, and Usage
Data source Relevant ATCS stage Case study-specific data
Field data for ϕii,ϕv, ϕvi, ϕvii Weekday turning movement
microsimulation counts at each intersection
Vehicle composition
Floating car travel time data
Saturation flow data (all data was
collected for PM peak period,
16:30–17:30)
ϕii Optimized TOD signal-control
timing parameters
Downloaded from ascelibrary.org by Universidad Nacional Autonoma de Mexico on 02/25/19. Copyright ASCE. For personal use only; all rights reserved.
ϕv, ϕvi, ϕvii Finetuned field ATCS
configurations
Bluetooth data ϕi–ϕvii Corridor travel time during
ATCS-off period
Corridor travel time during
ATCS-on period
Crowdsourced data ϕi–ϕvii Corridor travel time
Bluetooth Data
For the study corridor, Bluetooth monitoring stations (BMS) were
installed on the corridor to capture before-ATCS and after-ATCS
conditions. Bluetooth-enabled devices in the vehicles traveling
along the corridor serve as in-vehicle units that can be anonymously
recorded and location/time-stamped by the roadside BMS for the
purpose of vehicle identification. The BMS system calculates travel
time through the analysis of subsequent detections. To observe the
before traffic conditions for the study corridor, the ATCS was turned
off for one week from November 5–9, 2012. If the post-ATCS signal
timing parameters and the field conditions remain significantly
unchanged from the pre-ATCS condition, the before and off
evaluations provide statistically equivalent performance measures
(Kergaye et al. 2009). Since, to the best of our knowledge, no such
significant changes were noticed on the study corridor during the
ATCS-off period, this period can serve as the before-ATCS condi-
tion. From the next week onward, ATCS was turned back on for the
rest of the after-ATCS condition (where the ATCS is still running at
the time of this paper submission). This is called the ATCS-on
period for the Bluetooth data analysis purpose.
Crowdsourced Data
Deploying Bluetooth stations along the roadside incurs significant
costs and time. Crowdsourced data (which is a newer and emerging
method of measuring and assessing travel times on corridors)
provides an alternative. “Crowdsourcing involves leveraging the
combined intelligence, knowledge, or experience of a group of peo-
ple to answer a question, solve a problem, or manage a process”
(Michigan Department of Transportation and Center for Automo-
tive Research 2015). This technique leverages the network of con-
nected devices (probe devices), such as computers, smartphones,
and other mobile devices, to provide crowdsourced traffic data.
For the study purpose, probe data was obtained from commercially
available sources for the four-year period January 2011–December
2014. This data availability is one way to review speeds and travel
time data even years before a project begins.
Microsimulation Data
While travel time data is useful for characterizing main line perfor-
mance, side-street performance will need to be evaluated differently
to examine the trade-offs in operational policy goals. One useful
© ASCE 04017011-4
J. Transp. Eng., Part A: Systems, 2017, 143(5): 04017011
Data collection period Data usage in paper
May 2012 Simulating the before-ATCS and after-ATCS (ϕii
and ϕv) conditions in VISSIM
May 2012 Simulating the before-ATCS (ϕii) conditions in
VISSIM
October 2013 Simulating the after-ATCS (ϕv) conditions in
VISSIM
November 5–9, 2012 Used for ATCS-off (ϕiv) bluetooth data analyses
November 12–30, 2012 Used for ATCS-on (ϕv) bluetooth data analyses
4-year period from Monitoring impact of the policy changes and
January 2011 to adjustments to the ATCS on travel times (ϕvi)
December 2014 Long-term monitoring (ϕvii)
tool can be microsimulation, which gives the capability of con-
ducting a more detailed analysis of the ATCS than field studies
by observing every single vehicle. Once the ATCS has been turned
on and the system is in place, microsimulation models can be cali-
brated and validated to match different performance measures in
the field (Kergaye et al. 2010). To perform the microsimulation
analysis for this study, weekday volumes, turning movement counts
at each intersection, and vehicle compositions were collected for
the PM peak period (16:30–17:30) for the study corridor. Multiple
floating cars using multiple runs were used to collect the travel time
data during the before-ATCS period.
The saturation flow data for intersections that was initially col-
lected for a separate study was also used (Majeed et al. 2014). The
before-ATCS optimized field TOD signal timing parameters, and
the after-ATCS fine-tuned field ATCS signal configurations were
retained from the Traffic Engineering Department. The existing
operational data was used to develop VISSIM (version 5.4) micro-
simulation model. Detector locations (for both before-ATCS and
after-ATCS), signal phases, and other geometric conditions in the
simulation models were adjusted to match those in the field.
Results and Discussion
As discussed in the introduction, different analyses need to be con-
ducted at different times before, during, and after ATCS implemen-
tation for effective ATCS performance evaluations (Fig. 1). For the
before-ATCS period simulation techniques can be used but can be
challenging for analyzing the after-ATCS performance since it is
difficult to configure depending on the availability of ATCS configu-
ration files (often times these configuration files will only be available
from the vendor after the system has been turned on and fine-tuned in
the field). The results and discussion in this paper start with reviewing
various analyses during the implementation phase (reference Fig. 1,
callouts ϕiv and ϕv) for vendor and policy-maker feedback. Also, the
after analysis is examined where data and reports are more widely
available for review (reference Fig. 1, callouts ϕvi and ϕvii).
Corridor Travel Time Analysis of the Initial before/after
ATCS Implementation
Immediately after the ATCS is activated, corridor travel times can
be examined to identify any issues by direction and during different
J. Transp. Eng., Part A: Syst.
 time periods. For the turn-on period (reference Fig. 1, callout ϕiv),
Bluetooth data was analyzed during the first two-week period—one
week when ATCS was turned off and the immediately following
week when the ATCS system was turned back on (ideally, this
analysis could be conducted the week before the ATCS is turned
on without further interruption to the system). No notable weather
changes, special events, construction activities, or other anomalies
occurred along the corridor, which ostensibly provided good
ATCS-off and ATCS-on conditions. Fig. 3 shows the cumulative
distribution of the travel time for these two weeks (note that the
Downloaded from ascelibrary.org by Universidad Nacional Autonoma de Mexico on 02/25/19. Copyright ASCE. For personal use only; all rights reserved.
SCATS system referred to in the figures is the ATCS system evalu-
ated in this paper). The thinner red and green lines indicate the
Fig. 3. (Color) ATCS-off (one week) versus ATCS-on (one week)
(bluetooth data analysis for ϕiv): (a) calendar showing the SCATS eva-
luation time period using bluetooth data; (b) NB 07:30–08:30 AM peak
(bluetooth data); (c) SB 07:30–08:30 AM peak (bluetooth data); (d) NB
13:30–14:30 PM off-peak (bluetooth data); (e) SB 13:30–14:30 PM
off-peak (bluetooth data); (f) NB 16:30–17:30 PM peak (bluetooth
data); (g) SB 16:30–17:30 PM peak (bluetooth data)
© ASCE 04017011-5
J. Transp. Eng., Part A: Systems, 2017, 143(5): 04017011
cumulative distribution of each of the weekdays, while the bold
lines indicate the average of the five weekdays. Fig. 3(a) shows
the calendar indicating the time periods for which the analysis
is being carried out. Figs. 3(b, d, and f) show the plots for north-
bound (NB) AM peak, off-peak, and PM peak periods, whereas
Figs. 3(c, e, and g) show the plots for the southbound (SB) direction
for the similar periods. As can be seen from the graphs, the travel
times during the ATCS-on week as indicated by the green lines are
better than the travel times during the ATCS-off week as indicated
by the red lines. For example, referring to Fig. 3(d), the median
travel time during the ATCS-off period is 4∶17 min as compared
to the median travel time during the ATCS-on period of 4:06 min.
Fig. 4. (Color) ATCS-off (one week) versus ATCS-on (three weeks)
(bluetooth data analysis for ϕv): (a) calendar showing the SCATS eva-
luation time period using bluetooth data; (b) NB 07:30–08:30 AM peak
(bluetooth data); (c) SB 07:30–08:30 AM peak (bluetooth data); (d) NB
13:30–14:30 PM off-peak (bluetooth data); (e) SB 13:30–14:30 PM
off-peak (bluetooth data); (f) NB 16:30–17:30 PM peak (bluetooth
data); (g) SB 16:30–17:30 PM peak (bluetooth data)
J. Transp. Eng., Part A: Syst.
 Similar phenomena can be seen during the PM peak period, where
the median travel time for the ATCS-off period is better (smaller)
than the median travel time during the ATCS-on period. Interest-
ingly, while the AM peak travel times hardly show any improvement,
the improvement in the travel time as considered by the difference
between the travel times during ATCS-off and the ATCS-on period is
higher during the PM peak period in both northbound and south-
bound directions. This clearly indicates better ATCS performance
than the TOD signal plan. This analysis is performed based on travel
time data for only one ATCS-on week period. To verify the consis-
Downloaded from ascelibrary.org by Universidad Nacional Autonoma de Mexico on 02/25/19. Copyright ASCE. For personal use only; all rights reserved.
Corridor Travel Time Analysis of the Early Post-ATCS
Implementation Period
After the ATCS has been on and the vendor has made several early
changes, the first-month performance can be analyzed in a similar
fashion using the Bluetooth travel time data. Fig. 4 shows the sim-
ilar travel time cumulative distribution plots over a period of four
weeks in the month of November 2012. As shown in Fig. 4(a), the
first week is the ATCS-off period considered while the rest of three
weeks is the ATCS-on burn-in period (reference Fig. 1, callout ϕv).
The week-to-week travel time after the ATCS was turned on varied

tency and repeatability of performance, a similar ATCS-on analysis slightly but mainly shows an improvement during the ATCS-on
is further performed for a longer period. period. This indicates better ATCS performance during the PM

Fig. 5. (Color) Stock plot generation process using crowdsourced data: (a) INRIX TMCs and travel time calculations for every 1 min; (b) getting the
stock bar from the Pareto sorted travel time plots for 1 h; (c) 24 h travel times stock plot

© ASCE 04017011-6 J. Transp. Eng., Part A: Syst.

J. Transp. Eng., Part A: Systems, 2017, 143(5): 04017011

Downloaded from ascelibrary.org by Universidad Nacional Autonoma de Mexico on 02/25/19. Copyright ASCE. For personal use only; all rights reserved.

Fig. 6. (Color) Travel time distribution for before-and-after analysis (crowdsourced data for ϕvi and ϕvii): (a) NB 07:30–08:30 AM peak (crowd-
sourced data); (b) SB 07:30–08:30 AM peak (crowdsourced data); (c) NB 13:30–14:30 PM off-peak (crowdsourced data); (d) SB 13:30–14:30 PM
off-peak (crowdsourced data); (e) NB 16:30–17:30 PM peak (crowdsourced data); (f) SB 16:30–17:30 PM peak (crowdsourced data)

peak period in both directions than the other time periods studied
[Figs. 4(f and g)]. Examining the corridor travel times as shown in
Figs. 3 and 4 with a statistically robust approach is more useful than
floating car studies, which may not accurately characterize perfor-
mance over several hours of week-to-week periods. Corridor travel
time analysis during this period establishes the consistency of the
ATCS performance. If not, it would be advisable to extend this
analysis period until the consistency is established.
Long-Term before and after Analysis with
Crowdsourced Data
After the initial examination of corridor travel time performance, a
long-term monitoring approach should be adopted to check for sea-
sonal variability and other trends. Several studies have been per-
formed recently that compare various different methods and
techniques that have been used to collect travel time data (Remias
et al. 2013). These studies are essentially focused on finding the
technology that most reliably and accurately collects the data.
Although the procedural findings of these studies vary, they all
underscore the importance of higher sample size for higher accuracy
and reliability (Elefteriadou et al. 2014; Tahmasseby 2015). To verify
and validate the results from the Bluetooth data analysis, a similar
analysis was further performed using crowdsourced probe data.
The travel time along the corridor can be easily derived by con-
sidering the travel time of individual road segments that make up
the corridor. The data for each segment consists of the unique seg-
ment code (known as a traffic message channel or TMC in the fig-
ure), name, direction of travel, length of the segment, time stamp,
and the travel time. Fig. 5(a) shows how the corridor PQ is made up
of three segments Px, xy, and yQ. The total travel time for the cor-
ridor spanning from P to Q which is 5.50 min is the combined sum
of travel times of individual segments Px, xy, and yQ. This travel
time calculation is a facilities-based calculation representative of
the systemwide travel times as opposed to a user-based travel time
that would sum travel times for a probe user as they transverse
the corridor. The average travel time is calculated every minute.
Fig. 5(b) shows the cumulative distribution of travel time for a
60-min period. It also shows how a stock plot for a 1 h period
is generated using the cumulative distribution of the 60 travel times
that are time-stamped every minute in that hour. Distribution of
travel time over longer periods can be obtained by repeating this
process. Fig. 5(c) shows the stock plot of travel time for a 24 h
period. Such plots are extremely useful for travel time studies be-
cause they show the trend, pattern, and variability in travel time.
Fig. 6 shows the stock plot of the corridor travel time generated
using the crowdsourced data as explained in earlier sections. The
travel time is plotted over a four-year period for northbound and
southbound directions for different time periods. These plots show
how the travel time distribution changes with changes in the ATCS
settings. During the initial period from October 2012 to October
2013, the ATCS seems to perform better than the original TOD
signal plan. The same effect can also be observed in Fig. 7, which
shows the cumulative distributions of the travel times between the
periods of these changes. These graphs corroborate the results from
the Bluetooth data. The morning peak period hardly shows any im-
provement as depicted by the orange curves, which illustrate the
ATCS performance after installation. The evening off-peak and
peak period show improvements in travel time as was seen with
Bluetooth data. The trends shown by the crowdsourced data match
exactly to those shown by the Bluetooth data. The minor difference
in the magnitudes of travel time can be attributed to the difference
in data sampling rates as well as the sampling period. While Blue-
tooth data was collected for a month, the crowdsourced data ranges
for about 12 months after-ATCS but earlier to any changes made
to the system. It can also be seen that the ATCS performance
deteriorates when the local agency takes over and makes camera
upgrades in October 2013. The removal of lead/lag sequencing
further deteriorates the ATCS performance as indicated by the

© ASCE 04017011-7 J. Transp. Eng., Part A: Syst.

J. Transp. Eng., Part A: Systems, 2017, 143(5): 04017011


Downloaded from ascelibrary.org by Universidad Nacional Autonoma de Mexico on 02/25/19. Copyright ASCE. For personal use only; all rights reserved.
Fig. 7. (Color) Travel time cumulative distribution (crowdsourced data
for ϕvi and ϕvii): (a) NB 07:30–08:30 AM peak (crowdsourced data);
(b) SB 07:30–08:30 AM peak (crowdsourced data); (c) NB 13:30–
14:30 PM off-peak (crowdsourced data); (d) SB 13:30–14:30 PM
off-peak (crowdsourced data); (e) NB 16:30–17:30 PM peak (crowd-
sourced data); (f) SB 16:30–17:30 PM peak (crowdsourced data)
lighter blue lines. Finally, the adjustments in the marriage/divorce
settings further downgrade the ATCS performance. Note that each of
these changes resulted in higher main line travel times but are justified
in side-street delay improvements as discussed in the next section.
Before-After Analysis Using Microsimulation
The before-ATCS (reference Fig. 1, callout ϕii) model was cali-
brated manually using an iterative process of intuitively (based
on field data collected and observations) selecting the user-defined
parameters, such as driving behavior, lane change parameters, and
speed distributions, and by running the simulation and validating
the output against link volumes, turning movement counts, and
queues at each intersection. Synchro was used for volume balanc-
ing to avoid any directional data inconsistencies. Speed distribu-
tions and speed decisions were based on posted speed limits
and travel times. The model was validated by comparing the
modeled and field volumes, turning-movement counts at each sig-
nalized intersection, travel times between the signalized intersec-
tions, queues at selected intersections, and end-to-end corridor
travel times. Additionally, simulated travel times were checked
against data obtained from the Bluetooth and crowdsourced data
for further validation.
© ASCE 04017011-8
J. Transp. Eng., Part A: Systems, 2017, 143(5): 04017011
SCATSIM was used to run the after-ATCS (reference Fig. 1,
callout ϕv) simulation model. SCATSIM allows for the replication
of the SCATS adaptive control in the VISSIM environment. The
fine-tuned field ATCS signal configurations from the field (refer-
ence Fig. 1, callout ϕv) were used in the SCATSIM. No additional
calibration or validation efforts were made for the after-ATCS sim-
ulation model. As a result, the after-ATCS simulation model was
similar to the before-ATCS model in all respects except for the sig-
nal control. This allowed for the comparison of the same baseline
conditions under different signal control systems and hence objec-
tive evaluation of ATCS performance. It is important here to note
that this is an important aspect of the microsimulation technique. It
provides a truly objective comparison of the ATCS performance by
simulating the before-ATCS traffic conditions and the after-ATCS
signal controller, which is not possible with any other data sources.
Both networks before-ATCS and after-ATCS were simulated for
10 replications. Different random seeds were used over the 10 rep-
lications to capture the stochastic variation in traffic characteristics.
Warmup periods of 30 min were used to fill the network corridor to
the given traffic conditions prior to recording performance mea-
sures. Simulation durations of 60 min were used to evaluate the
network performance (Dowling et al. 2004).
The simulation results for the corridor travel time are shown in
Fig. 8 and are consistent with the field travel time observations.
Main line as well as networkwide performances were analyzed sep-
arately as shown in the two sections of Fig. 8. The main line per-
formance includes the end-to-end corridor performance and
excludes the side-streets performance. The networkwide travel time
includes side-streets performance in addition to the main line per-
formance. Travel times in the main line northbound direction show
improvement under ATCS control. However, in the southbound
main line direction, travel time increases under ATCS control.
However, the networkwide performance shows that the total net-
work travel time as measured by the sum of travel times of all indi-
vidual vehicles that traveled the network has significantly improved
after-ATCS. This can be seen in Fig. 8(c).
While the main line performance shows mixed results, the network-
wide performance clearly indicates consistent and significant improved
performance. At this stage, this contrast can be attributed to the only
uncommon factor between the two performance measures, the side
streets. To validate this notion, the queue lengths and nodewise delays
per vehicle were examined for the main line and side streets separately.
Figs. 9(a and b) show the average of the left-turn and through
movement queue lengths formed at every intersection in the north-
bound and southbound direction, respectively. Although the aver-
age queue lengths at every intersection in the northbound direction
[Fig. 9(a)] are shorter under the ATCS regime, the average queue
lengths in the southbound direction are a little longer [Fig. 9(b)]
with ATCS. However, the queue lengths on the side streets show
about 58% improvement in queue lengths. The same trade-off is
seen from the nodewise average delay per vehicle calculated as
the average of the left-turn and through movement for every inter-
section. Although the delays on the main line in either direction are
smaller, the side-street delays show greater improvements with
ATCS. Figs. 9(d–f) show the average delay per vehicle for every
intersection. The average side-street delay per vehicle improves by
about 45%. This validates the notion that the improvement in the
networkwide performance is mainly contributed by the improved
side-streets performance. This is important feedback for system op-
erators as they continuously try to balance operational performance
goals between main line travel time and delay on the side streets.
This analysis can be further extended to find other interesting re-
sults. While various data sources of travel time data can be used to
characterize the corridor performance before-ATCS or after-ATCS,
J. Transp. Eng., Part A: Syst.
Downloaded from ascelibrary.org by Universidad Nacional Autonoma de Mexico on 02/25/19. Copyright ASCE. For personal use only; all rights reserved.

Fig. 8. (Color) Travel time (simulation): (a) NB travel time; (b) SB travel time; (c) total travel time

Fig. 9. (Color) Comparison of main-line and side-street performances (simulation): (a) NB average queue lengths; (b) SB average queue lengths;
(c) EB and WB average queue lengths; (d) NB average delay per vehicle; (e) SB average delay per vehicle; (f) EB andWB average delay per vehicle

a well-calibrated and validated simulation model can still be used to
analyze additional measures that characterize ATCS performance.
Conclusions
ATCS systems are an advanced tool that can improve traffic
operations. The ATCS implementation case study in this paper
was performed to compare the arterial performance before, during,
and after ATCS deployment by examining the travel time data and
to demonstrate different analyses that can be conducted at different
phases of the ATCS implementation. The methodologies discussed
will help transportation engineers, planners, and policymakers to
monitor the system performance and improvements at any point
of the ATCS implementation process. As policies and timings are

© ASCE 04017011-9 J. Transp. Eng., Part A: Syst.

J. Transp. Eng., Part A: Systems, 2017, 143(5): 04017011

 adjusted, this type of feedback is an effective way to continuously
monitor the ATCS and quantify benefits. The study yielded the fol-
lowing general conclusions/recommendations:
1. Different sources of travel time data can be used to characterize
the corridor performance before-ATCS or after-ATCS. This
study shows that the corridor travel times measured by Blue-
tooth data as well as crowdsourced data are very close and
showed similar trends. The minor differences in the travel time
arise out of different sampling rates and periods.
2. Early data collection before the ATCS installation begins is re-
commended. Newer crowdsourced data can provide a long-term
Downloaded from ascelibrary.org by Universidad Nacional Autonoma de Mexico on 02/25/19. Copyright ASCE. For personal use only; all rights reserved.
look at main line corridor travel times over the life (including
before and during installation) of an ATCS.
3. A well-calibrated and validated simulation model can deliver
results close enough to the field conditions. The advantage of
simulation technique is to compare the same baseline traffic
conditions under different signal control systems, which may
not be possible to evaluate with other data sources.
4. Depending on operational policies and goals, performance mea-
sures other than main line travel time may need to be considered.
Although corridor travel time is a typical traditional way to as-
sess arterial improvements, other measures need to be studied
for comprehensive evaluation of new, complex systems, such
as ATCS. This paper leveraged calibrated simulation models
to gain detailed insight of side-street and networkwide perfor-
mance measures to supplement the information gleaned from
the travel time data. Microsimulation can be used to examine
longer-term impacts of policy changes and system adjustment
throughout various stages of ATCS deployment. To do so, how-
ever, requires additional data collection and model calibration/
validation.
5. As every technique has its own limitations, multiple evaluation
techniques should be used to improve evaluations and resulting
decision making. Engineers and other persons involved with the
implementation will need to think out their own detailed analy-
sis plan to reflect project-specific timelines, performance goals,
and other considerations.
Considering these conclusions, it is worthwhile to note that
while data is becoming more available, using the data for effective
evaluation and analysis still remains an evolving standard that may
tend to be project specific. Future work in this area should consider
using ATCS-specific reporting data and also traffic controller data
collected over time to assess performance changes.
Acknowledgments
This work was supported by the Alabama Department of Transpor-
tation. The contents of this paper reflect the views of the authors,
who are responsible for the facts and the accuracy of the data pre-
sented herein, and do not necessarily reflect the official views or
policies of the sponsoring organizations. These contents do not
constitute a standard, specification, or regulation.
References
Dowling, R., Skabardonis, A., and Alexiadis, V. (2004). “Traffic analysis
toolbox. Volume III: Guidelines for applying traffic microsimulation
modeling software.” FHWA-HRT-04-040, FHWA, U.S. Dept. of Trans-
portation, McLean, VA.
© ASCE 04017011-10
J. Transp. Eng., Part A: Systems, 2017, 143(5): 04017011
Elefteriadou, L., Kondyli, A., and George, B. (2014). “Comparison of
methods for measuring travel time at Florida freeways and arterials.”
Transportation Research Center, Univ. of Florida, Gainesville, FL.
Gord & Associates. (2007). “Adaptive versus traditional traffic control
systems—A field-based comparative assessment.” 〈https://www
.pinellascounty.org/mpo/ITS/Final-Evaluation-Paper-Field-Studies-040907
.pdf〉 (Jun. 17, 2015).
Grant, M., Bauer, J., Plaskon, T., and Mason, J. (2010). “Advancing met-
ropolitan planning for operations: An objectives-driven, performance
based approach: A guidebook.” FHWA-HOP-10-026, FHWA, Federal
Transit Administration, U.S. Dept. of Transportation. Washington, DC.
Hunter, M., Roe, M., and Wu, S. (2010). “Hardware-in-the-loop simulation
evaluation of adaptive signal control.” Transp. Res. Rec., 2192,
167–176.
Hutton, J., Bokenkroger, C., and Meyer, M. (2010). “Evaluation of an adap-
tive traffic signal system: Route 291 in Lee’s Summit, Missouri.” 〈http://
trid.trb.org/view.aspx?id=916407〉 (Jun. 17, 2015).
Kergaye, C., Stevanovic, A., and Martin, P. (2010). “Comparative evalu-
ation of adaptive traffic control system assessments through field
and microsimulation.” J. Intell. Transp. Syst., 14(2), 109–124.
Kergaye, C., Stevanovic, A., and Martin, P. T. (2009). “Comparison of
before-after versus off-on adaptive traffic control evaluations.” Transp.
Res. Rec., 2128, 192–201.
Majeed, A. A., Zephaniah, S. O., Mehta, G., and Jones, S. (2014). “Field-
based saturation headway model for planning level applications.” Int. J.
Traffic Transp. Eng., 3(5), 207–215.
Martin, P. T., and Stevanovic, A. (2008). “Adaptive signal control V—
SCATS evaluation in Park City, Utah.” 〈http://www.mountain-plains
.org/pubs/pdf/MPC08-200.pdf〉 (Sep. 17, 2015).
MDOT (Michigan Department of Transportation) and Center for Automo-
tive Research. (2015). “Crowdsourcing transportation systems data.”
MDOT REQ. NO. 1259, Lansing, MI.
National Transportation Operations Coalition. (2012). “2012 National traf-
fic signal report card.” 〈http://www.ite.org/reportcard/TechnicalReport
.pdf〉 (Jul. 11, 2015).
Peters, J. M., McCoy, J., and Bertini, R. (2007). “Evaluating an adaptive
signal control system in Gresham.” Western ITE, 61(5), 1–8.
Remias, S., et al. (2013). “Performance characterization of arterial traffic
flow with probe vehicle data.” Transp. Res. Rec., 2380, 10–21.
SCATSIM [Computer software]. Roads & Maritime Services (former Roads
and Traffic Authority), Liverpool, NSW, Australia.
Schrank, D., Eisele, B., and Lomax, T. (2012). “TTI’s 2012 urban mobility
report.” Texas A&M Transportation Institute, Texas A&M Univ. Sys-
tem, TX, 〈http://mobility.tamu.edu〉 (Jun. 16, 2015).
Schrank, D., Eisele, B., Lomax, T., and Bak, J. (2015). “2015 urban mobil-
ity scorecard.” Texas A&M Transportation Institute and INRIX, Inc.
〈https://mobility.tamu.edu/ums/report/〉 (Aug. 28, 2015).
Shelby, S. G., Bullock, D. M., Gettman, D., Ghaman, R. S., Sabra, Z. A.,
and Soyke, N. (2008). “An overview and performance evaluation
of ACS lite—A low cost adaptive signal control system.” 〈https://w3.
usa.siemens.com/mobility/us/en/urban-mobility/road-solutions/adaptive
-software/Documents/ACS_Lite_Overview_TRB_2008_CD.pdf〉 (Feb. 27,
2016).
SRF Consulting Group, Inc. (2000). “AUSCI—Adaptive urban signal
control and integration.” SRF No. 0942089.8, Minnesota Dept. of
Transportation, Roseville, MN.
Stevanovic, A. (2010). “Adaptive traffic control systems: Domestic and for-
eign state of practice.” NCHRP Synthesis 403, NCHRP, Washington,
DC.
Tahmasseby, S. (2015). “Traffic data: Bluetooth sensors vs. crowdsourcing—
A comparative study to calculate travel time reliability in calgary, alberta,
canada.” J. Traffic Transp. Eng., 3, 63–79.
Tian, Z., Ohene, F., and Hu, P. (2011). “Arterial performance evaluation on
an adaptive traffic signal control system.” Procedia—Soc. Behav. Sci.,
16, 230–239.
VISSIM 5.40 [Computer software]. Planung Transport Verkehr AG (PTV),
Karlsruhe, Germany.
J. Transp. Eng., Part A: Syst.