Content uploaded by Abdullah Almaatouq
Author content
All content in this area was uploaded by Abdullah Almaatouq on Feb 27, 2015
Content may be subject to copyright.
The City Browser: Utilizing Massive Call Data to Infer City
Mobility Dynamics
Fahad Alhasoun
Center for Computational
Engineering
Massachusetts Institute of
Technology
Cambridge MA, USA
fha@mit.edu
Abdullah Almaatouq
Center for Complex
Engineering Systems at
KACST and MIT
Riyadh, Saudi Arabia
amaatouq@mit.edu
Kael Greco
Senseable City Lab
Massachusetts Institute of
Technology
Cambridge MA, USA
kael@mit.edu
Riccardo Campari
Senseable City Lab
Massachusetts Institute of
Technology
Cambridge MA, USA
campari@mit.edu
Anas Alfaris
Center for Complex
Engineering Systems at
KACST and MIT
Riyadh, Saudi Arabia
anas@mit.edu
Carlo Ratti
Senseable City Lab
Massachusetts Institute of
Technology
Cambridge MA, USA
ratti@mit.edu
ABSTRACT
This paper presents the City Browser, a tool developed to analyze
the complexities underlying human mobility at the city scale. The
tool uses data generated from mobile phones as a proxy to provide
several insights with regards to the commuting patterns of the pop-
ulation within the bounds of a city. The three major components
of the browser are the data warehouse, modules and algorithm,
and the visualization interface. The modules and algorithm com-
ponent utilizes Call Detail Records (CDRs) stored within the data
warehouse to infer mobility patterns that are then communicated
through the visualization interface. The modules and algorithm
component consists of four modules: the spatial-temporal decom-
position module, the home/work capturing module, the community
detection module, and the flow estimation module. The visualiza-
tion interface manages the output of each module to provide a com-
prehensive view of a city’s mobility dynamics over varying time
scales. A case study is presented on the city of Riyadh in Saudi
Arabia, where the browser was developed to better understand city
mobility patterns.
Categories and Subject Descriptors
H.2.8 [Database Applications]: Data Mining; H.2.8 [Database
Applications]: Spatial Databases and GIS; H.4 [Information Sys-
tems Applications]: Miscellaneous
General Terms
Urban Computing, Data Analytics, City Science
Copyright is held by the author/owner(s).
UrbComp’14, August 24, 2014, New York, NY, USA.
Keywords
Call Detail Records, Mobile Applications, Urban Analysis
1. INTRODUCTION
Cities today house over 50 percent of world’s population, con-
suming 60-80 percent of global energy and emitting almost 75 per-
cent of greenhouse gases [18]. Some have suggested that almost
70 percent of world’s population will reside in cities by 2050 [18].
With the rapid urban population growth, cities’ infrastructures are
being strained to the point of becoming a major hindrance to so-
cioeconomic activity. Left unaddressed, the problem threatens to
weigh down the return on investment from public projects being
constructed throughout cities and adversely affect the quality of life
of all residents.
Understanding the complexities underlying the emerging behav-
iors of human travel patterns on the city level is essential toward
making informed decision-making pertaining to urban transporta-
tion infrastructures [2]. Traditional methods of assessing the social
demand on transportation are expensive and take longer periods of
time to conduct [10, 19, 21]. Such assessments are usually in the
form of surveys with considerably small sample sizes compared to
the total population of a city. Furthermore, such methods lack the
accuracy and resolution in time to provide fine-grained analysis of
human travel with precise time resolution.
New road counter technologies such as pressure tubes, inductive
loops and other traffic counting techniques allowed for counting
travelers with a finer time resolution; however, the drawback is the
spatial resolution of such techniques. They are usually highly local
and capture activity in a specific point in space that is miniscule
with respect to the city as a whole [16]. Therefore, such techniques
suffer from an inability to provide a holistic overview of the status
of the system. In addition, deploying new traffic counting technolo-
gies can be extremely expensive when considering the mega-cities
in the world.
An alternative approach toward capturing the social demand is
by using data generated from mobile phones to model and un-
derstand the behavior of human mobility [21]. Data pertaining to
mobile phone usage can be gathered at different levels within the
GSM network. Telecom companies usually do not keep track of all
the data traffic running across their networks; however, they store
certain information for billing purposes and network development.
The Call Detail Records, often referred to as CDRs, are one type
of information telecom companies keep for billing purposes. Ev-
ery time a user makes a phone call, sends a text message uses the
Internet and even passively when the mobile communicates to the
cellular network access points, the mobile network keeps a record
of their usage information and location in the CDRs [11]. There-
fore, such big data set can be utilized as a proxy to understand the
social demand on transportation infrastructures.
The motivation behind developing the browser is derived from
the demand of a tool that provides fine-grained analysis of the com-
plexity of human travel within cites. The approach takes advantage
of the existing built infrastructures to sense the mobility of people
eliminating the financial and temporal burdens of traditional meth-
ods. The outcomes of the tool will assist both planners and the
public in understanding the complexities of human mobility within
their cities.
In this paper we will present the City Mobility Browser, a tool
that facilitates a simplified understanding of human mobility across
a city. The paper is divided into four sections: Section 2 describes
existing methods and approaches, Section 3 presents the method-
ology of the browser, Section 4 describes the general architecture
of the system, Section 5 describes each component of the tool in
detail, and Section 7 presents results of the case study of city of
Riyadh in Saudi Arabia. The contributions of this paper can be
summarized into the following two points:
•We propose an architecture that combines several known tech-
niques for data collection, storage and analysis in one frame-
work in a meaningful context to develop the “City Browser”,
that can aid in simplifying the complexity of human mobility
across a city.
•We examine the usefulness of the system through a case study
of Riyadh, Saudi Arabia. The case study contained 100 mil-
lion real mobile phone activity and demonstrates the process
of analyzing massive amount of data and through visualiza-
tion, distilling the bits into actionable insights.
2. BACKGROUND
Several research activities have been investigating approaches to-
wards modeling and understanding mobility demand within cities.
Traditional methods of demand modeling inferred the collective be-
havior of demand on transportation infrastructures through house-
hold or road surveys to gather information about user’s behavior.
Another approach has been to use theoretical models to estimate
the number of trips and their directionality based on land use mod-
els. These approaches can be unreliable and can have financial and
temporal costs. Today and with the emergence of pervasive tech-
nologies around the world, research started investigating human
behavior through data gathered from mobile phones [8, 9, 12, 13].
Varying approaches have used the data as a proxy to better under-
stand human mobility. The focus on human mobility ranges from
decomposing the data onto the different dimensions to gain insights
into behavioral patterns by applying algorithms and processes on
models built on the data. Research investigating the dimensional-
ity of the data includes work on utilizing the spatial decomposi-
tion of aggregate activity to understand the dynamics of cities and
universal patterns of human mobility [3, 9]. On the other hand,
researchers have developed techniques to gain more insights from
the data by creating algorithms capturing more of the hidden pat-
terns [7, 12,22]. For example, researchers have been modeling the
social network based on the data captured from users’ interactions
to better understand whether the composition of social communi-
ties is correlating with the geographical constraints [17]. Another
approach was to capture users’ trips from the data set and aggregate
trips to get insight on the flows of people around the city towards
understanding the dynamics of flows of people [4,21]. Such under-
standing can help identify flawed urban planning in cities [23].
3. METHODOLOGY
The objective of the browser is to provide an understanding of
the complexity underlying human mobility within a city. The browser
will capture the dynamics of the distribution of the population to in-
vestigate aspects pertaining to flows of people as well as the struc-
ture of the community. Investigating population localization dy-
namics provides information pertaining to emerging zones with
higher population densities; certain dense zones emerge on daily
basis like commercial areas on weekdays while others emerge as
consequence of events that are not of periodic nature. The browser
will investigate whether the formation of periodic dense zones has
an influence on the segregating of the population of the city into
communities. On the other hand, it will provide information about
how the city interacts with events in terms of population commut-
ing flows.
The approach towards simplifying the complexity of human mo-
bility is staged into four steps. Starting with step 1, the browser de-
composes population distribution across the spatial dimension on
a time resolution of a day capturing the emergence of dense zones
(see Subsection 3.1). Step 2 then analyzes each individual in the
CDRs to capture their home/work locations (see Subsection 3.2).
Step 3 as explained in subsection 3.3 investigates the formation of
communities within cities as a result of their home/work choices.
Step 4 estimates people flows within the city within a day time scale
(see Subsection 3.4).
3.1 Spatial-Temporal Decomposition
The first phase of the methodology decomposes the population
over the spatial dimension of the city on the day scale; it will cap-
ture time series information of densities of people at every zone
with time granularity in minutes. The technique quantifies the mag-
nitudes of mobile user activities within the defined time window,
generating time series data for user activity densities for each zone
covered by a cell tower. Observing densities with such fine time
granularity provides fine grained detail on the emergence of such
populated zones by identifying when, where and how fast different
dense zones emerge.
3.2 Home/Work Places Capturing
The second phase takes a larger time granularity spanning weeks
to capture residential and business areas. The approach towards
that is by identifying locations where users spend most of their time
during day and night (i.e. home/work locations) across a sufficient
time interval. Aggregating the number of users spending most of
their times over a particular location captures zones that are emerg-
ing as a result of daily routine activities like regular business areas
and schools.
3.3 Community Detection
To better understand the influence of where people live and work,
this phase investigates the formation of segregated communities
based on their home and work locations. The formation of a mobil-
ity community within the population indicates that there is a subset
of the population traveling within confined bounds of the city and
tend not to cross those bounds (i.e. a neighborhood or group of
neighborhoods). Such analysis can provide insights on the level of
heterogeneity of trips’ sources and destinations.
3.4 Flows Estimation
To better understand daily commuting within a city, this phase
captures flows within the city through the origin destination esti-
mation algorithm. The algorithm captures trips generated by users
around the day and then aggregates the flows of people on a spec-
ified time window. The results of the origin destination estimation
algorithm will provide information about how dense zones emerge
in terms of the source of the population visiting those zones.
4. GENERAL ARCHITECTURE
The general architecture of the browser is composed of three ma-
jor components; data warehouse, modules and algorithms, and the
visualization interface. The data warehouse contains the needed
data for the modules and algorithms to produce insights and infor-
mation visualized through the visualization interface. The general
architecture is shown in the figure 1. The data warehouse contains
data pertaining to human mobile phone usage as well as GIS infor-
mation of the city and traffic counts. There are four major mod-
ules residing within the modules and algorithms component that
are spatial-temporal decomposition module, home/work capturing
module, community detection module and flows estimation mod-
ule. Finally the visualization interface takes the results produced
by the modules and algorithms together with GIS information of
the city to provide a comprehensive dynamic view of human mo-
bility within a city.
Figure 1: City Browser general architecture
"details of the implementation of the architecture"
5. COMPONENTS
The City Browser is decomposed into components following the
general architecture described in section 4. This section will pro-
vide the details of each component. The breakdown of the browser
into components is to allow for a more scalable, modular and sim-
pler architecture for development. Each of the components is de-
scribes below.
5.1 Data Warehouse
The data warehouse houses several datasets containing informa-
tion of the structure of the city as well as the dynamics of it. It
contains a geospatial database of the city including the lookup ta-
ble of the locations of the cell towers for the purpose of mapping
mobile phone activity to locations. In addition, it contains infor-
mation of the time series mobile phone usage data as well as traffic
counts.
The major part of the data warehouse is mobile phone billing
information, also known as Call Detail Records (CDRs), which
are records that telecom companies usually keep for the purpose of
generating bills for customers. The CDRs are generated by mobile
switching centers (MSCs) within GSM networks and go through
several processing methods to be usable by telecom providers. The
CDRs are finally structured in a table-like format, withholding in-
formation about phone activity details. Each entry in the CDRs
table is a record representing an activity generated by a user. Every
time a user makes a phone call, sends a text message or accesses the
Internet, the CDRs keeps a record of the cell tower that was used to
facilitate activity. In addition, the data warehouse contains a lookup
table for cell tower geospatial information where each cell tower is
mapped to its coordinates (i.e. latitude and longitude). Each record
within the databases is referred to as an activity and is described
by time t, user uand cell tower cand represented as a(t, c, u). For
each user, the dataset contains a series of activities captured and are
represented in this paper as:
Au={a0, a1, a2, ...an|u=ua0=ua1=ua2=.... =uan}
where a0is an activity record and ua0is the user generating activ-
ity a0. The data warehouse also contains traffic volume counts at
specific points on the road network. Traffic counts are usually taken
for a defined period of time where pressure tubes are placed on cer-
tain links to count the number of times vehicles pass across them.
Furthermore, information about the geometry of the road network
is housed within the data warehouse as a spatial database. The road
network spatial database contains information about the geometry
of roads such as number of lanes, category, length and speed limit.
5.2 Modules and Algorithms
The Modules and Algorithms component is composed of four
components: spatial-temporal decomposition module, home/work
capturing module, flow estimation module, and community detec-
tion module. Each of the components is described below.
5.2.1 Spatial-Temporal Decomposition Module
The first step toward understanding the dynamics of a city on the
day scale is to look at the dynamics of population densities across
the city through aggregate user activities for each cell tower. This
module breaks down the total activities of users on both the spatial
and temporal dimensions. A similar approach was developed in [4].
For each cell tower within the city, the module generates a time
series data for activity levels for a specified time granularity ∆t. To
capture the collective behavior of the population across the city, the
module captures the aggregate activity level of users at every cell
tower ciwithin the city. The aggregate phone activity level denoted
AL(ci,∆t)at cell tower cifor a time window ∆tis computed as
follows from the dataset.
AL(ci,∆t) = X
c∈ci,t∈∆t
a(c, t, u)
Where a(c, t, u)is an activity generated through cell tower cat
time t. Each time series data for every location cigives insights on
the nature of the zone where the cell tower resides in terms of its
use. For example, work areas within cities are expected to have a
higher density of activity during work hours compared to residen-
tial areas. The module also provides insight into collective pop-
ulation behavioral characteristics showing when the city becomes
alive in the morning. It also captures information on how users are
interacting with events in terms of localization or behavior of ser-
vice usage. The objective of developing this module is to provide a
holistic overview of the change in population densities across space
and time.
5.2.2 Capturing Home/Work Places Module
Expanding the time interval of the analysis, this module captures
work zones as well as residentail zones. This is essentially cap-
turing places where the majority of daytime calls are as a proxy to
work locations. First, we segregate activity records on two time
windows to capture most visited zones at daytime versus nighttime
for a particular user u. Activities that would hold potential work
locations are separated in a set as:
dayu={a0, a1, a2, ...an|u=ua0=ua1=ua2=.... =uan
∧tai∈daytime}
Where a0is an activity record, ua0is the user generating activity
a0and taiis time tag of activity ai. Similarly, nightuis obtained
with the same logic for nighttime activity. Then, workulocation
for user uis chosen to be the most occuring location in dayuand
the same applies to homeuas it is chosen to be the most occuring
location in nightu.
After determining the workuand homeufor each user. The
aggregation of the resulting zones where users spend most of their
times during the day and night identifies dense zones that pertain
to business/residential areas since the module considers larger time
granularity for the analysis. Thus, this module quantifies the extent
to which a zone is considered as residentail/business zone.
5.2.3 Community Detection Module
Following on the output of section 5.2.2, this module will in-
vestigate whether there are groups within the population forming
communities that have similar home and work locations. The mod-
ule begins with the city-wide network of connected zones G(N, E)
where Nis the set of cell towers within the city representing the
zones and Eis composed of weighted directed edges defined as
the number of users who have a particular home/work pair, respec-
tively, in the zones corresponding to the starting and terminating
nodes. The adjacency matrix Aof the discussed network is as fol-
lows:
A=
w0,0w0,1· · ·
w1,0w1,1· · ·
.
.
..
.
.· · ·
wm,1wm,2
...
Where w0,1is the number of users having their homeuas c0
and workuas c1. The algorithm then uses a modularity optimiza-
tion scheme, such that sets of nodes are clustered in a way that
minimizes internal arc disruption [5, 14]. Each resulting commu-
nity represents an area where a large fraction of users are mostly
located during the day and night.
Modularity is a standard objective function in the field of com-
munity detection; it measures how well a partition of network nodes
into communities reflects the characteristics of the underlying net-
work (in our case the commuting flow among zones). The ratio-
nale behind modularity is that a group of nodes with connections
mostly directed towards its own members represent a community
with higher modularity while a set of nodes with intra-community
connections is what we would expect by randomly rewiring all the
links.
Communities resulting from modularity optimization of telecom-
munication data have been empirically shown to be representative
of the actual social and administrative boundaries at the level of
whole countries [7].
In the case of a city, we went further and studied communities at
the level of the neighborhood. The interesting results we obtained
are discussed in Section 7.
5.2.4 Flows Estimation Module
To capture the directionality and mobility of the population across
the city, the browser houses an algorithm that provides information
about the collective behavior of human mobility through mining
mobile phone activity. The module of estimating the aggregate
flows of people across the city from the CDRs is a three step al-
gorithm that has the CDRs as inputs and the aggregation of flows
of people between locations at every time window ∆tas its result
(i.e. Origin Destination matrix). A similar approach was developed
in [4]. The module starts by arranging data on a user level and con-
sidering each of their displacements as a potential trip. After that,
the resulting potential trips go through a filtration process that fil-
ters out noise in the data from the potential trips generated. Finally
the last step aggregates the resulting trips on both the spatial and
temporal dimensions to generate an origin-destination matrix based
on the provided time slice of interest.
The first step in the algorithm looks at phone activities on a user
level and gathers all activities generated for each user sorted in time
as follows.
Au={a0, a1, a2, ...an|u=ua0=ua1=ua2=.... =uan
∧ta0< ta1< ta2....tan}
Where Auis the set of all activities generated by user u,uaiis
the user generating the activity aiand taiis the time tag of activ-
ity ai. Every consecutive records belonging to the same user are
merged into pairs of location records with their associated times
representing a potential displacement of the user. The set of dis-
placements of a user are represented as given by:
Du={(cai, cai+1 , tai, tai+2 )|a0, a1,∈Au}
Where Duis the set of all potential displacements of user u,
caiis the cell tower facilitating the activity ai,taiis the time tag
of activity aiand uaiis the user generating activity ai. The set of
potential displacement considers each successive user activity a po-
tential trip though this includes noisy data such as users who did not
change their locations between the successive activities but where
nevertheless served by different nearby cell towers, a phenomena
referred to as localization error. In order to capture user trips in
which a displacement actually occurs, we apply further filtering on
the set of potential displacements Du. The goal of the filtering pro-
cess is to eliminate all captured pairs of location records that are
considered as noise in terms of trip-capturing. The filtration pro-
cess eliminates all records that are considered as localization error,
have very long time intervals or no movement detected. Entries
in the data that corresponds to localization error are filtered out by
eliminating all trips that are less than a specified distance of the
maximum distance between any neighboring cell towers within an
urban setting. Given any two neighboring cell towers that caiand
caj, each element within Dumust satisfy the below predicate.
distance(cai, cai+1 )> max[distance(cai, caj)]
Where disance(cai, cai+1 )is the distance between the towers
caiand cai+1 . The filter eliminates potential displacements having
a distance larger than that of the maximum distance between any
two neighboring towers in the city. In addition, each pair of records
satisfy tai+1 −tai> α, where taiis the time tag of activity ai.
That is a time difference between consecutive activity records being
more than a threshold is filtered out of the set of displacements Du
for the purpose of reducing the uncertainty in capturing the actual
departure and arrival times for trips.
The result of the filtering process is the set of displacements ¯
Du
containing all pairs of locations where movement was detected and
reasonable time duration for the trip was captured. After that, the
final step towards the generation of OD matrices is to aggregate the
trips according to the specified time slice into the origin destination
matrix given by:
OD(∆t) =
0T0,1T0,2· · ·
T1,00T1,2· · ·
T2,0T2,10· · ·
.
.
..
.
.....
.
.
Where each element Ti,j gives the number of trips captured be-
tween cito cjduring the time slice ∆t. The value of Ti,j is com-
puted by:
Ti,j (∆t) = X¯
Du(can, can+1 , tan, tan+1 )
Where can∈i, can+1 ∈jand tan+1 −tan∈∆t. Thus, Ti,j
quantifies the flows from zone ito zone jduring the time window
∆t.
6. VISUALIZATION INTERFACE
The visualization component shows the results of the modules
and algorithms on two time scales depending on the nature of their
outputs. It will visualize population density distribution and ma-
jor flows of people across the city dynamically over the span of a
day while on longer time scales it will show a static map of the
communities forming around the analysis of dense zones.
The visualization will start by showing the spatial-temporal de-
composition of the population over the scale of a day. A dynamic
visualization with time granularity of 15-minutes will capture pop-
ulation density variations across the day and night. The browser
shows mobile activity over a dynamic period of time broken up
into 15-minute intervals as shown in figure 3. This visualization
presents a rotatable, scalable map onto which a shifting, three-
dimensional grid is superimposed to show locational agglomera-
tions of cellphone activity. Grid sectors will rise and fall, and
brighten and fade as people move across the city using their mo-
bile devices.
On the same scale of a day, the visualization components shows
the directionality of human mobility through the output of flow es-
timation module as well as the car counts stored in the data set.
Major flows within the city showing the aggregate behavior of com-
muting around the day are visualized with a time window of 15-
minutes. The component visualizes the generation of trips on each
time slice by as an arc that rose from originating to terminating cell
tower. As shown in figure 6, each arc embodies a variable num-
ber of trips, and to illustrate this we altered its thickness and height
in correspondence to the intensity of activity along that route (on
a logarithmic scale). The arcs are drawn over the same city base
geography, on top of the social interaction mesh from above, in an
effort to reveal unseen connections between the two results. In ad-
dition, car counts were built into the visualization as half-spheres
placed at their respective intersections. Each sphere changes shape
and color at an hourly rhythm in line with the measured volume.
On the longer time scale and towards visualizing the output of
the community detection module, the visualization interface over-
lays the community network over the spatial dimension of the city
to show if there are correlations between the formation of commu-
nities and the urban fabric of the city. Nodes represent zones of
the city and arcs represent groups of people spending most of their
times across the day/night between connected nodes. The commu-
nity detection module provides the set of nodes that belong to the
same community. To visualize the output of the community detec-
tion algorithm, nodes belonging to the same community are col-
ored with the same color as shown in figure 5. Thus, areas where
sub communities spend most of their time during the day and night
are bounded within zones of the same color.
7. CASE STUDY
Over the past decade, Saudi Arabia has taken strong steps to-
wards developing a diversified economy. Specifically on enhanc-
ing its Information and Communication Technology (ICT) infras-
tructure [1]. Today, Saudi Arabia has one of the highest Internet
penetration percentages in the gulf area with current penetration at
14.7 million. It is ranked among the highest countries worldwide in
mobile penetration rates with 188% of the population possess mo-
bile phones [6]. The high penetration rate of mobile in Saudi Ara-
bia make it an ideal candidate for utilizing the Call Detail Records
(CDRs) as in situ sensors for human mobility.
The City Browser was implemented for the Urban Transporta-
tion System (UTS), a system developed to provide city planners
with insights with regards to the mobility of the population. The
project started with gathering information related to the structure
of the city as well as the dynamics of the population. The data
gathered includes Records CDRs spanning a period of the month
of December, a spatial database of the road network of Riyadh city
and traffic counts data on different points within the city. Currently,
the data is housed within the data warehouse where several modules
and algorithms are using it to generate insights on the dynamics of
the city.
7.1 Data Description
Our dataset consists of one full month of records for the entire
country of Saudi Arabia, with 3 billion mobile activities to over 10
thousands unique cell towers, provided by a single carrier. Each
record contains an anonymized user ID, the type of activity (i.e.,
SMS, MMS, call, data etc), the cell tower facilitating the service,
duration if its a phone call, and time stamp of the activity. Each
cell tower id is spatially mapped to its latitude and longitude. For
privacy concerns, user id information were completely anonymized
at the telecom operator side.
Previous studies [9, 15] have shown that human communication
patterns are highly heterogeneous; where some users use their mo-
bile phone much more frequently than others. The characteristics
(a)
(b) (c)
Figure 2: Communication patterns in the CDRs Dataset. Fig 2a
shows the Empirical Cumulative Distribution Function (ECDF) of
the activities duration. We find that almost 75% of the users con-
duct activities that last for 70 seconds or less. Fig 2b shows the
statistical distribution of the number of communication records
generated by the users for a single day. Fig 2c shows the inter-
event time distribution P r(∆t)of calling activity, where ∆tis the
time elapsed between consecutive communication records (outgo-
ing phone calls and SMS) for the same user.
of the dynamics of individual communication activity obtained in
Fig 2 supports such hypothesis.
7.2 City Spatial-temporal Decomposition
The first step towards understanding the data in the city of Riyadh
is to decompose cellular activity on the spatial and temporal dimen-
sions. The visualization in figure 3 shows cellular activity through
color, transparency, and height (in logarithmic scale) gridded across
the metropolitan expanse of Riyadh. As opposed to seeing the cell
towers as discrete points in the city, we show network traffic in-
terpolated over a 100 by 100 grid. In this sense, each grid cell is
assigned an intensity based on its distance to surrounding anten-
nas and their activity levels using a Gaussian smoothing function.
The temporal activity is interpolated in a similar manner, show-
ing smooth transitions between each time-slice in the dataset.The
city’s downtown core quickly becomes clouded in smog of network
activity early in the morning that hangs over region for the entire
day. Clear sub centers emerge that follow construction density, and
these sub-centers appear to be partitioned by the roadway network
itself.
Figure 3: Spatial-Temporal Decomposition out for a single time
slice. The figure demonstrates the time-cumulative spatial mobile
activity conducted between 9:45am to 10:00am.
The city’s shifting activity profile also highlights a rich tempo-
ral signature of communication that is all Riyadh’s own. Watching
the oscillations of the activity landscape, we see that Riyadh comes
alive at around 6:15am. We also see strong regional delineation:
the residential neighborhoods to the southwest and northeast of the
downtown core come alive well before the rest of the city, and expe-
rience the strongest inter-hour fluctuations throughout the course of
the day. Finally, we see some peculiar discontinuities in aggregate
talk throughout the day almost as if all phone traffic was suddenly
halved at strange intervals.
7.3 Capturing Home/Work Places
A fundamental quality of mobility behavior is to analyze the
emergence of zones with higher densities along a wider time gran-
ularity to understand the distribution of residential and business
zones. Expanding our time intervals to capture broader day and
night variation we can begin to differentiate dense business areas
and schools versus dense residential neighborhoods.
Figure 4: Dense work zones during the day versus home locations
during the night. We observe high day-densities at the periphery
where major universities are located.
The map in Figure 4 highlights the discrepancy between the purely
day zones shifting towards the red color and the purely night dense
zones shifting towards blue color, showing some mono-centrically
clustered day hotspots that follow the overall spatial logic of the
city. At the periphery we also see a number of universities show
up strongly as day locations. Lastly, we see high agglomerations
of residences to both the south and east of the city, with smaller
pockets scattered throughout.
7.4 Detecting Mobility Communities
The work/home dense zones visualizations shown in section 7.3
point to an organizational logic of the city. Conceptualizing the
totality of day/night commutes as a city-wide mobility network,
we can conceivably break this network into sub-communities by
applying a regional delineation algorithm.
Figure 5: Community Detection Module results plotted by Latitude
and Longitude on the map of Riyadh. We find support to the com-
monly held belief that heavily trafficked streets, on many levels, are
instruments of segregation and control.
By overlaying the results of the community detection module on
geography of the city (see Figure 5), a number of interesting rela-
tionships are revealed between the detected communities and the
built form of the city. Most strikingly, the resulting clusters closely
correlate to the main arterials of city’s roadway infrastructure. Mo-
bility communities seem to be partitioned by the street network it-
self, underscoring the city’s dependence on highway infrastructure,
while also supporting the commonly held belief that heavily traf-
ficked streets, on many levels, are instruments of segregation and
control, or, perhaps more optimistically: good streets make good
neighbors.
7.5 Flow Estimation
The approach toward understanding flows that contribute dense-
zone emergence on smaller time granularity unveils rich informa-
tion pertaining to the sources of dense zones as well as the distri-
bution of flow over time. By collecting and filtering each user’s
mobile activity as sequence of cell tower locations and then aggre-
gating collective users’ trips, we are able to estimate flows in terms
of origins and destinations of trips. We’ve observed that these esti-
mated flows contributed to the emergence of high density zones in
the city of Riyadh; however this approach includes the added bene-
fit of capturing travel demand at highly dynamic time slices ranging
from seasonal variations to hourly fluctuations. Such a high tem-
poral resolution has the potential to transform our understanding of
urban mobility [20].
Figure 6: The extracted Origin Destination (OD) matrix across
Riyadh at the time slice of 9:30-9:45am. The height of the line
corresponds to the number of trips between a specific OD.
The resulting dynamic maps held a striking similarity to the lo-
cal intuition of vehicular flows across the city (see Figure 6). Over-
all flows correspond quite closely to the underlying street network.
Most notably, Figure 6 shows intense activity along the city’s main
arterials; King Fahd Road and the Northern and Eastern Ring roads.
This agrees with the local community’s subjective understanding of
commute patterns across the city. But to further validate our results,
we compared them against the best ground-truth measurements of
roadway activity: car count volumes captured by pressure-tube sen-
sors placed at multiple intersection across the city.
8. SUMMARY AND FUTURE WORK
In this paper, we have presented a new tool addressing the com-
plexity of city human mobility and showed its application to the
city of Riyadh the capital of Saudi Arabia through the UTS project.
The project developed the Riyadh Mobility Browser by implement-
ing several modules that mined data generated from mobile phones
to provide a coherent understanding of the dynamics of the interac-
tion between its social structure and transportation infrastructures.
At the current stage, the browser is built to work with historical
data and thus would provide an after-the-fact analysis and does not
allow for the parsing and analysis of the data in real time. A poten-
tial future work would be investigating the possibility of enabling
the browser to parse such big data in real time through establishing
a live connection of data feed with GSM network operators.
The city mobility browser synthesizes and extends existing al-
gorithms to provide a holistic decomposition of the complexity of
mobility across multiple dimensions. Although the browser cap-
tures the dynamics of the demand on transportation, it does not
map the demand over the road network of the city.
We also acknowledge that some of the explanations and conclu-
sions proposed in this work might lack rigorous validations and this
is due to the nature of the CDRs where it lacks sufficient granular-
ity in space and time. Spatially, the data is mapped to the locations
of cell towers and not the exact locations of users and therefore the
coordinates of cell towers are used as a proxy to the exact loca-
tions of users. Temporally, users have a bursty phone usage behav-
ior where activities are clustered around different times of the day
rather than spread out around the day to enable a more comprehen-
sive understanding of mobility in this case. However, we believe
that our analysis of human mobility can describe well the current
trends and phenomenon of human mobility and can be leveraged in
planning the city and transportation operations.
The visualizations provided by the tool give a dynamic qualita-
tive understanding of the spatial attributes of the city as well as its
population directionality across different times of the day. The city
mobility browser is envisioned to be a tool that can provide plan-
ners, engineers and the public with an easy to understand analysis
while capturing fine grained details about the city. Future work
could also enable the visualization interface to provide quantitative
analysis and a better understanding of emerging patterns.
9. REFERENCES
[1] A. Almaatouq, F. Alhasoun, R. Campari, and A. Alfaris. The
influence of social norms on synchronous versus
asynchronous communication technologies. In Proceedings
of the 1st ACM International Workshop on Personal Data
Meets Distributed Multimedia, PDM ’13, pages 39–42, New
York, NY, USA, 2013. ACM.
[2] M. Batty, K. Axhausen, F. Giannotti, A. Pozdnoukhov,
A. Bazzani, M. Wachowicz, G. Ouzounis, and Y. Portugali.
Smart cities of the future. The European Physical Journal
Special Topics, 214(1):481–518, 2012.
[3] D. Brockmann, L. Hufnagel, and T. Geisel. The scaling laws
of human travel. Nature, 439(7075):462–465, Jan. 2006.
[4] F. Calabrese, M. Colonna, P. Lovisolo, D. Parata, and
C. Ratti. Real-time urban monitoring using cell phones: A
case study in rome. IEEE Transactions on Intelligent
Transportation Systems, 12(1):141–151, 2011.
[5] A. Clauset, M. E. J. Newman, and C. Moore. Finding
community structure in very large networks. Physical Review
E, 70:066111, 2004.
[6] Communications and I. Commission. Ict indicators, q2-2012,
2012.
[7] Y.-A. de Montjoye, C. A. Hidalgo, M. Verleysen, and V. D.
Blondel. Unique in the Crowd: The privacy bounds of human
mobility. Scientific Reports, 3, Mar. 2013.
[8] N. Eagle and A. (Sandy) Pentland. Reality mining: Sensing
complex social systems. Personal Ubiquitous Comput.,
10(4):255–268, Mar. 2006.
[9] M. C. Gonzalez, C. A. Hidalgo, and A.-L. Barabasi.
Understanding individual human mobility patterns. Nature,
453(7196):779–782, June 2008.
[10] J. Herrera, D. Work, R. Herring, X. Ban, Q. Jacobson, and
A. Bayen. Evaluation of traffic data obtained via
GPS-enabled mobile phones: The Mobile Century field
experiment. Transportation Research Part C, 18(4):568–583,
August 2010.
[11] S. Jiang, G. A. Fiore, Y. Yang, J. Ferreira, Jr., E. Frazzoli,
and M. C. González. A review of urban computing for
mobile phone traces: Current methods, challenges and
opportunities. In Proceedings of the 2Nd ACM SIGKDD
International Workshop on Urban Computing, UrbComp
’13, pages 2:1–2:9, New York, NY, USA, 2013. ACM.
[12] S. Jiang, J. F. Jr, and M. González. Discovering urban
spatial-temporal structure from human activity patterns. In
Proceedings of the ACM SIGKDD International Workshop
on Urban Computing, pages 95–102, 2012.
[13] J. R. Kwapisz, G. M. Weiss, and S. A. Moore. Activity
recognition using cell phone accelerometers. SIGKDD
Explor. Newsl., 12(2):74–82, Mar. 2011.
[14] M. E. Newman. Modularity and community structure in
networks. Proc Natl Acad Sci U S A, 103(23):8577–8582,
June 2006.
[15] J. Onnela, J. Saramaki, J. Hyvonen, G. Szabo, D. Lazer,
K. Kaski, J. Kertesz, and A. L. Barabasi. Structure and tie
strengths in mobile communication networks. Proc. Natl.
Acad. Sci. USA, 104(18):7332–7336, 2007.
[16] S. Peeta and P. a. Zhang. Counting device selection and
reliability: Synthesis study. Technical report, Purdue
University.
[17] C. Ratti, S. Sobolevsky, F. Calabrese, C. Andris, J. Reades,
M. Martino, R. Claxton, and S. H. Strogatz. Redrawing the
map of great britain from a network of human interactions.
PLoS ONE, 5(12):e14248, 12 2010.
[18] P. Rode and R. Burdett. Cities: investing in energy and
resource efficiency. In United Nations Envrionment
Programme, (corp. ed.) Towards a Green Economy:
Pathways to Sustainable Development and Poverty
Eradication, pages 453–492. United Nations Environment
Programme, 2011.
[19] W. Shen and L. Wynter. Real-time traffic prediction using
GPS data with low sampling rates: A hybrid approach. In
91st Transportation Research Board Annual Meeting,
number 12-1692, Washington, D.C., January 2012.
[20] J. L. Toole, S. Colak, F. Alhasoun, A. Evsukoff, and M. C.
Gonzalez. The path most travelled: Mining road usage
patterns from massive call data. arXiv preprint
arXiv:1403.0636, 2014.
[21] P. Wang, T. Hunter, A. M. Bayen, K. Schechtner, and M. C.
González. Understanding Road Usage Patterns in Urban
Areas. Scientific Reports, 2, Dec. 2012.
[22] J. Yuan, Y. Zheng, and X. Xie. Discovering regions of
different functions in a city using human mobility and pois.
In Proceedings of the 18th ACM SIGKDD international
conference on Knowledge discovery and data mining, pages
186–194. ACM, 2012.
[23] Y. Zheng, Y. Liu, J. Yuan, and X. Xie. Urban computing with
taxicabs. In Proceedings of the 13th international conference
on Ubiquitous computing, pages 89–98. ACM, 2011.
