26 ISE Magazine | www.iise.org/ISEmagazine
The U.S. Transportation Security Administration (TSA) is responsible for security at near-
ly 440 federalized airports. TSAs approximately 50,000 transportation security officers
(TSOs) screened over 2,000,000 passengers per day departing on over 25,000 daily flights
before the COVID-19 pandemic. While air travel dropped, levels had rebounded to more
than 1,000,000 daily passengers as of February 2021.
Planning operations to minimize wait times at security screening checkpoints (SSCP)
while ensuring flight safety with limited operational resources and rapidly changing conditions pres-
ents a challenge, but one that ISE tools are well-designed to conquer. The DHS Center for Acceler-
ating Operational Efficiency (CAOE) at Arizona State University has teamed with TSA to address
this challenge.
TSA has standard passenger and baggage screening procedures and equipment at airports for en-
suring flight safety. The supply of security officers that operate the checkpoints are set through an
employee bidding process with limits specified by Congress. With the equipment infrastructure and
workforce level fixed, the operational goal is to match available screening resources with the screen-
ing demand caused by passengers arriving for flights. The goal is to ensure that passenger wait times
do not exceed a desired time, such as 10 minutes, at security checkpoints. The basic question to be
determined is the number of travel document checker (TDC) stations and carry-on baggage screen-
ing lanes to have open at each checkpoint at each point of time during the day.
Passenger arrival estimation
The first step in matching screening capacity to demand is to predict demand, i.e., the number of
passengers arriving for screening in each time segment. While this may sound simple, the reality is
more complicated due to data availability and inherent randomness.
Individual airlines set their flight schedules and equipment (airplanes) for each route. These plans
change frequently, and for competitive reasons, airlines may not want to share their full flight sched-
ules and capability well in advance. Some seats are only sold at the last minute. Flight delays and
gate assignments may change in real time. Employees at airport retail outlets may be changing shifts
within the secured area. Moreover, the airport business model is evolving to include more shopping,
dining and entertainment facilities for travelers.
All these factors complicate an estimate of the number of individuals arriving at a security check-
point for screening at a particular time. To model the demand, 10-minute time segments were chosen
to capture the dynamically changing arrival rates and ability to open and close screening lanes with-
out incurring excessive stochasticity and computation.
A basic mechanistic model is used to capture the quantiable factors that drive arrivals. The model
includes flight schedules, early arrival behavior (the distribution of how long prior to a flight’s sched-
uled departure that passengers arrive for screening), equipment capacity (available seats), load factor
(the percent of available seats occupied), the percent of passengers originating at the airport and the
probability a flight will depart from gates controlled by the specific security checkpoint.
T
Improving the efficiency of
airport security screening
checkpoints
Passenger data calculated to speed up preflight processes
By Ronald G. Askin and Jorge A. Sefair
May 2021 | ISE Magazine 27
28 ISE Magazine | www.iise.org/ISEmagazine
Improving the efficiency of airport security screening checkpoints
The model for the number of passengers arriving at secu-
rity screening checkpoints g for time interval t on day j then
becomes Equation 1:
This is where F
j
is the set of flights for day j; r
g
is the propor-
tion of flights assigned to SSCP g; c
k
is seat capacity of flight k;
l
k
is the load factor (proportion of sold seats); o
k
is the propor-
tion of passengers originating at this airport; and a
kt
is the pro-
portion of passengers of flight k that are expected to arrive for
screening in period t. This latter factor is computed from a set
of earliness arrival curves empirically fitted based on the flight
departure time and type of flight (domestic or international).
The mechanistic model described in Equation 1 uses avail-
able information from commercial or governmental sources.
An alternative to the historical capacity, load and originating
factor combination is to use the reports TSA receives from air-
lines on cleared passengers. Unfortunately, that data can vary
significantly up to the day of departure, making it difficult to
use for advance planning.
The causal model captures the underlying basics of arriv-
als. However, it fails to adjust for factors such as when airport
workers may be going through screening during shift changes
or the impact of seasonal loads and holidays. Likewise, some
of the available data is aggregated. Load factors, for instance,
are not available by flight destination or day and time of flight
as might be desired.
To include this information, the model is expanded by su-
pervised machine learning to adjust the arrival numbers us-
ing historical data from the past years, months or weeks (or a
combination) for the same hour of the day and day of the week
at each checkpoint. Holiday adjustments can also be included.
For instance, Thanksgiving travel can be matched. The modi-
fied arrival estimates are constrained to be within 25% of the
original estimates but adjustment parameters are fitted within
the training (historical) set by regression to minimize the dif-
ference between the predicted values and the observed values
of passengers screened.
The resulting (trained) adjustment parameters are used in
the forecasting. Figure 1 shows a sample of the predicted ar-
rivals for an SSCP.
Estimating queue lengths and wait times
Passenger screening is primarily a two-stage process. The pas-
sengers present their travel documentation for ID verication
to a travel document checker, who then passes the individual
through to the second stage. The second stage entails passen-
gers placing their baggage on a conveyor for X-ray inspec-
tion, then passing through a walk-through metal detector or
advanced imaging technology machine. After inspection, the
passengers collect their belongings and depart.
In a few cases, a passenger is selected for secondary inspec-
tion involving residue testing, detailed baggage check or other
activity that will add an additional delay. Typically, there is
limited space between the document checker and scanning
stages, and thus TDC stations may become blocked.
A simple input-output flow conservation model can be
used to estimate queue
lengths and wait times.
Using the notation de-
ned in the table below,
the basic discrete pro-
cessing model becomes
as shown in Equation
2. The expression re-
flects that the number
of passengers processed
in time interval t is the
smaller of the num-
ber of passengers in the
queue at the start of the
(10-minute) period plus
those that arrive to that
stage and the processing
capacity of that stage. In
practice, an adjustment
is made to the expres-
sions in Equation 2 to
account for the fact that
the actual process is sto-
FIGURE 1
Predicted arrivals
The forecast passenger arrivals in 10-minute intervals throughout a single day at an airport security screening
checkpoint.
May 2021 | ISE Magazine 29
chastic, and thus the queue is unlikely to be 0 even if capacity
exceeds arrivals. Thus, the second term in the min expressions
are adjusted in accordance with queueing theory.
Notation for security screening checkpoint throughput
analysis
C is the set of TDC workstations.
S is the set of X-ray lane screening lines.
j(i) is the set of X-ray lines associated with TDC workstation
i.
Qit is the number of passengers at station i at the start of
period t.
B
j(i)
is the maximum passenger capacity in between TDC i
and screening lane j(i).
c
it
is the number of passengers processed at station i in period
t.
is the number of passengers arriving for screening during
period t.
• is the processing capacity (service rate) in passengers per
period for station i in period t.
This equation states that given the processing quantities at
each stage, queue lengths are updated each period by: Ending
queue length = Starting queue length + Arrivals – Departures.
For the travel document checkers, arrivals are determined by
Equation 1. For baggage screening, arrivals are determined by
the passengers that pass through the document checker stands.
The TDCs typically have more capacity than the screening
lanes and thus are throttled back when the available space fills
by the last term in the first line of Equation 2.
Waiting times are also estimated for passengers arriving in
each time period. Arriving passengers are assumed to enter at
the back of the queue present at the start of the period. Then
using the c
it
processing values, wait times are computed by
nding the number of whole plus fractional periods required
until the cumulative c
it
values equal the queue at the time the
passenger arrived.
In this way, an arriving passenger has a wait time at the
document checking station, then an additional wait time based
on the queue length when arriving to the screening lane. The
reciprocal of the service rate at the station gives the service
time, and that is added to the waiting time to arrive at the
passenger’s throughput time. Similar calculations are made for
general boarding and precheck passengers. Figure 2 shows a
sample of the predicted wait times, i.e., the total time from a
passenger’s arrival at the security screening checkpoint until
departure to the gate.
A user spreadsheet toolkit
The models are implemented in an Excel spreadsheet tool that
can be used by operational planners. Separate worksheets are
used to input basic parameters such as the number of document
checker and screening lanes available, the processing rate of
each, the number of security officers required to operate each
station, the allowable queue space at each screening lane and
the number of officers avail-
able at each 10-minute inter-
val. The number of available
ofcers can be computed
directly from the employee
database that contains each
TSO and his or her schedule,
including availability due to
sick leave, training or other
factors. A separate worksheet
contains flight information
including departure time and
capacity. The system will au-
tomatically recommend a
conguration that seeks to
minimize excessive waits.
Alternatively, users can in-
put their own conguration,
which allows for quick what-
if analyses on which docu-
ment checker and screening
lanes to open, and when to
do so. In the base output, the
FIGURE 2
Total wait time (general boarding)
Sample plot of total wait times across the 10-minute periods of a day.
30 ISE Magazine | www.iise.org/ISEmagazine
Improving the efficiency of airport security screening checkpoints
recommended conguration assumes ample security officers
are available (nding an optimal conguration with limited
ofcers is described in the next section.) With the congura-
tion specified and the anticipated arrivals, the system computes
the expected queue lengths for each 10-minute period and
the expected throughput time, both total and by stage, for a
passenger arriving at the start of the period. Both tabular and
graphical outputs are provided.
A sample output is shown in Figure 3 for a securing screen-
ing checkpoint based on the Phoenix Sky Harbor airport.
Transportation security officer allocation
With passenger arrivals estimated for each time interval and a
model to transform arrivals and service system design into wait
times, the final step is to try to optimize the service system.
In this case, that translates into selecting a conguration of
the number of travel document checkers and screening lanes
open at each security checkpoint throughout the day to meet
a performance goal for wait time. Although we focus on SSCP
congurations, the model is easily extended to include other
airport screening systems such as checked luggage inspection.
In that case, employee qualifications are added since not all
security officers are qualified for both luggage and passenger
inspection.
In all cases, though, restrictions exist based on the available
infrastructure, TSOs available by time of day, movement times
to change checkpoints or open a lane, and organizational rules
such as the minimum time a lane must be open once opened
and how frequently an officer can be moved. A preprocessor
reads from the employee database to extract the set of TSOs
scheduled for the day on each shift, accounting for individuals
on vacations, sick leave and other “losses.” Shifts have dened
start and end times such as from 6-10 a.m. Break times are not
included in the decision model and left to the discretion of the
shift supervisor. Thus, the output of this preprocessing is the
exact number of security officers scheduled to be available at
the airport terminal in each time period.
Several optimization models have been constructed based
on the desire to minimize the number of officers needed for
a specific performance measure, such as maximum wait time,
or to minimize the maximum queue length subject to a limit
on the number of TSOs available each period. Factors such as
the need for at least one male and one female officer at each
checkpoint are not directly modeled but left to the detailed
implementation by the shift supervisor.
FIGURE 3
Sample spreadsheet tabular output
A sample output is shown for a securing screening checkpoint based on the Phoenix Sky Harbor airport.
Learn more about the center’s
security process efforts
Authors Ronald G. Askin and Jorge A. Sefair are members of
the research staff of the Center for Accelerating Operational
Efficiency (CAOE), led by Arizona State University (caoe.
asu.edu). The center’s goal is to develop and apply analytical
tools and technologies to enhance the U.S. Department of
Homeland Security’s planning, information sharing and
real-time decision-making. The research aims to improve
efficiency and security at national borders, ports and airports
through better prediction and response to emergencies.
You can watch a video explaining the airport security effort
on YouTube at link.iise.org/caoe_airports.
May 2021 | ISE Magazine 31
The primary decision variable is x
ilt
, a binary variable indi-
cating whether conguration l is selected for SSCP i in time
interval t. A conguration defines the number of document
checker and screening lanes open at a specific precheck or stan-
dard checkpoint screening area.
A description of the constraints in a typical model run is
included in the table below. In the table, the goal is to mini-
mize the maximum queue length occurring at any security
checkpoint area during the day subject to a limited budget on
how many additional officers can be called in to work. Set-
ting the limit to 0 would indicate no additional workers are
allowed, but in practice, floating TSOs can often be called in
or reassigned from other tasks, if needed, to meet performance
goals. Airport space, as well as optics, is a reason for minimiz-
ing queue lengths but wait times can also be minimized by a
suitable adjustment of the model. These may differ since the
time to empty the queue will depend on the conguration
(number of document checkers and screening lanes open).
Base model for screening conguration optimization
Objective: Minimize Q
max
Subject to:
1. Select exactly one conguration for each SSCP area each
time interval.
2. Set service rate at each SSCP area equal to that of selected
conguration.
3. Number of TSOs required for selected conguration;
TSOs available for each period.
4. Number of additional TSOs called in.
5. Queue length by area and time interval.
6. TDC passengers served.
7. Screened passengers.
8. Processed passengers.
9. TDC passengers served.
10. Ending queue = Starting queue + Arrivals – Number
processed for each area each period.
11. Queue length Lanes open x buffer space per lane for
each screening lane each period.
12. Drastic (unimplementable) conguration changes be-
tween successive periods are disallowed.
13. Conguration indicators are binary; queue lengths and
passengers served are continuous.
The model described in this table ensures that ample TSOs
are available either through the base human resources schedule
of individuals or by calling in extra flex workers. Using the
conguration selected, passenger arrival predictions and phys-
ical layout (space), the queue dynamics are tracked. Drastic
conguration changes are avoided but the model still does not
account for any time to move officers between security check-
points. Instead, a second optimization model that includes ac-
tual time to reassign officers between locations is used to check
implementation feasibility.
If the proposed conguration is found to be infeasible due
to TSO move time, that conguration sequence (or a subset in
the sequence) is restricted and the model is resolved. That pro-
cess can continue until a feasible solution is found. In practice,
this is implemented using a branch-and-cut approach with a
separation algorithm that either certifies the feasibility of the
suggested security officer movements or constructs a cut to
prevent them.
Screening procedures
The project discussed to this point represents one phase of the
work performed to increase the operational efciency of TSAs
security checkpoints and reduce passenger wait times. Related
work includes development and use of simulation models to
verify the analytical models and to investigate the impact of
various policy changes. Interested readers may contact the au-
thors for further details.
For instance, simulation studies have shown that adding an
extra security officer to the screening lane to help passengers
place luggage on the X-ray conveyor in the proper manner can
measurably increase the lane capacity. The impact of technol-
ogy changes such as automated travel document checkers or
alternate X-ray lane congurations can also be readily assessed
with the simulation models.
Note: This material is based upon work supported by the
U.S. Department of Homeland Security under Grant Award
Number 17STQAC00001-02-00. The views and conclusions
contained in this document are those of the authors and should
not be interpreted as necessarily representing the official poli-
cies, either expressed or implied, of the U.S. Department of
Homeland Security. The authors acknowledge TSA for its col-
laboration in this work, especially James Huffman and Scot
Thaxton as well as students Troy Munson, Casin Corallo, Ber-
kin Arici, Chao Wang and Girish Jampani Hanumantha.
Ronald G. Askin is a professor of industrial engineering at Arizona
State University. His teaching and research focus on the application of
industrial engineering and operations research techniques for produc-
tion/capacity planning and facility logistics. He is an IISE Fellow,
recipient of the Albert G. Holzman Distinguished Educator Award
and serves as the IISE senior vice president for publications. Contact
him at ron.askin@asu.edu.
Jorge A. Sefair is an assistant professor of industrial engineering at
Arizona State University. His research interests include network op-
timization, multistage optimization and large-scale optimization. In
particular, he is motivated by applications of operations research in envi-
ronmental planning, public policy and urban systems. His research has
been interdisciplinary, having published academic works with colleagues
from a variety of fields, including civil engineering, public health, ecol-
ogy, biology and economics.