MODELLING DYNAMIC PATTERNS USING

MOBILE DATA
Suhad Faisal Behadili1, Cyrille Bertelle1 and Loay E. George2
1

Normandie Univ, LITIS, FR CNRS 3638, ISCN, ULH, Le Havre, France

[email protected]
[email protected]
2

Baghdad University, Computer Science Department, Baghdad, Iraq

[email protected]

ABSTRACT
Understanding, modeling and simulating human mobility among urban regions is very challengeable
effort. It is very important in rescue situations for many kinds of events, either in the indoor events like
evacuation of buildings or outdoor ones like public assemblies, community evacuation, in exigency
situations there are several incidents could be happened, the overcrowding causes injuries and death
cases, which are emerged during emergency situations, as well as it serves urban planning and smart
cities. The aim of this study is to explore the characteristics of human mobility patterns, and model them
mathematically depending on inter-event time and traveled distances (displacements) parameters by using
CDRs (Call Detailed Records) during Armada festival in France. However, the results of the numerical
simulation endorse the other studies findings in that the most of real systems patterns are almost follows an
exponential distribution. In the future the mobility patterns could be classified according (work or off)
days, and the radius of gyration could be considered as effective parameter in modelling human mobility.

KEYWORDS
Modelling, Armada, Probability Distribution, Inter-event time, Displacements

1. INTRODUCTION
The purpose of simulation analysis is to acquire and analyze the results in well conceptual vision,
in order to give high indications for decision makers, pivoting on two events kinds, the discrete
events and frequent (continuous) events. In order to explore the mobility characteristics, eventual
real effects conditions, and actions of a specified system, they should be modeled mathematically
[1, 2, 3, 4, 5]. However, the most important and difficult issue in modeling and simulating any
system is the determination of the probability distribution and parameters to model the
uncertainty of the system input variables. Many researches used CDRs to study the collective and
individual human mobility (Gonzalez et al., 2008), the segmentation of urban spaces (Reades et
al. 2009), understand of social events (Calabrese et al., 2010). (Phithakkitnukoon et al., 2010)
suggested the activity-aware map, using the user mobile to uncover the dynamic of inhabitants,
for urban planning, and transportation purposes. (Ratti et al., 2010) deal with large
telecommunication database for Great Britain in order to explore human interactions, and
emphasized on the highly interactions correlation with administrative regions [10].
David C. Wyld et al. (Eds) : CCSEA, CLOUD, DKMP, SEA, SIPRO - 2016
pp. 2530, 2016. CS & IT-CSCP 2016
DOI : 10.5121/csit.2016.60203

26

Computer Science & Information Technology (CS & IT)

2. DATA SET
The case study data is Call detail Records of mobile phone, composed of 51,958,652 CDRs,
represent entry records of 615,712 subscribers, for the period starting from 4 (Friday)-15
(Tuesday) of July in 2008. However, it contains individuals occurrence in discrete (irrelevant)
mode only, means that any mobile individual activity is recorded at (start/end) time, but there is a
lack of information, which is supposed to indicate the individual's occurrence during inactive
case (mobility without any mobile phone activity). There is no data meanwhile the mobile phone
is idle, i.e. inactive or doesn't make any communication activities neither calls nor SMS activities.
As well as, the available spatial data is only of the towers (X,Y) coordinates, hence it would be
considered to estimate individual's transitions from position to another (from tower to tower),
hence the positions would be determined approximately with regarding to the tower coverage
(signal strength). With regarding to the non-deterministic & discrete nature of this data, therefore
the collective behavior would be the effective approach to be analyzed and simulated. Since, each
individual could be disappeared for a while from the DB, which makes individual tracing is
unworthy, without significant indications on the people behavior in the city [3].

3. DESCRIBING INDIVIDUALS' ACTIVITIES

The communications activities have heterogeneous nature, since the individuals are varied in
their usage of mobile phone, which are ranged between (rarely-frequently) usages during specific
period. The individuals are grouped according to their total activities. However, probability of
waiting time (inter-event time) T of each two consecutive activities has been computed for each
individual, so the individuals would be grouped with regarding to their activities. Computing the
probability function is to get the system universal behavior (pattern), according to consecutive
inter-event times, where most life systems are modeled by exponential law [1, 6, 7, 5, 8]. So, the
distribution of the average inert-event time is estimated by exponential distribution law as in
equation (1), its histogram in figure 1. The demonstration in figure 2, reveals that the longer
waiting times are characterizes the individuals of fewer activities.
=

Figure 1: Histogram of Activities in Average of Inter-Event Time

Computer Science & Information Technology (CS & IT)

27

4. INTER-EVENT TIME OBSERVATIONS

The modeling and simulation process starts by computing probability model, which is modeled
the data and simulate the responses of the model, using one of the well-known functions, the
probability density function PDF [3, 4, 9]. The main parameter in this study is the mean of interevent time, where the means of samples of the distribution are drawn as exponential distribution,
then compute the mean of all means. In order to understand the behavior of sample means from
the exponential distribution, to get the universal system pattern (general population law). The
computations are done as follows:
1. Manipulate all 12 days data, each day independently due to daily regular patterns, for all
users (individuals) in spatio-temporal manner. Then eliminate the individuals of only one
occurrence in the CDRs, since they didn't have significant indication on mobility.
2. Sorting the data by time, in order to have the real sequence of positions transitions of
individuals trajectories. Classify the data according to individuals' activities (sampling), by
computing inter-event time t (waiting time), where it is the time elapsed between each
successive activities of each individual, which is ranged between 15-1440 minutes, this
sampling is done according to logical intuition, since 15 minutes is the minimum time that
can give mobility indication, and the 1440 minutes (24 hours) could be considered as the
highest elapsed time to travel inside the observed region.
3. Compute the inter-event time of all individuals T, then compute (average inter- event
time) of all individuals, then classify (min, max) samples according to activities score
(activities densities), then compute. However, / is the average inter-event time
of the all individuals (whole population).
4. Compute exponential distribution probability for each day, then for the whole days as in
figure 2 to identify the universal population pattern law, then to show the approximation of all
12 days curves with the curve of their average values as in figure 3.

Figure 2: Waiting time distribution P(t) of mobile activities, where t is spent time between
each two successive activities, legend symbols are used to distinguish the individuals' groups
according to their activities ratio, for whole population during whole period, activities in
average t.

28

Computer Science & Information Technology (CS & IT)

Figure 3: Waiting time for the 12 days curves with the curve of their average.

Exponential distribution (probability distribution) in equation (1) is capable of modeling the

events happened randomly over time. In this case it describes the inter-event time, and the
average of inter-event time of individuals' activities [1, 2, 3, 9]. Hence, the cutoff distribution is
determined by the maximum observed inter-event time (significant parameters), which the
individual can wait to make any mobile activities, where it is t=1431 min. The resulted law as in
equation (1) is computed according to algorithm of complexity + 2 + . The
distribution shows that short inter-event times have higher probability than long ones, and the 12
days have similar patterns of activities. As well as, by following the same manner mentioned
above, in order to compute the displacement statistics of all the individuals during the whole
observed period, the displacement probability distribution is computed. However, the r is the
traveled distance between each two successive activities during time of the range 20-1440
minutes, and the P(r) is the displacement distribution, the investigated distances would be
limited by the maximum distance that may be traveled by individuals in the (time intervals).
Hence, the cutoff distribution is determined by the maximum observed distance, which individual
can travel is r=7.229515 Km along the day hours, since the maximum time slice couldn't exceed
24 hours with regards to observed region. The displacements distribution is approximated by
power low as in equation (2). The resulted distribution shows as in figure 4 that the displacements
are clear in 10 m, whereas decreased after this value, the r is ranged between (0-1) Km.
=

Well, the traveled distance (displacements) distributions of the 12 days curves with the curve of
their average are demonstrated in figure 5, they have almost identical patterns. Hence, the
obtained averages of the waiting time distribution are max T = 1431 minute and min T =0
minute, whereas the max r =7.229515e+04 meter and min r =0 meter.

Computer Science & Information Technology (CS & IT)

29

Figure 4: Distances probability distribution (displacements) P(r) for waiting times (inter-event
times) for one day, cutoff distribution is determined by the maximum distance traveled by
individuals for specific t .

Figure 5: Displacements for the 12 days curves with the curve of their average.

5. CONCLUSIONS
The great attending recently towards dealing with catastrophic disasters, cities planning, diseases
spreading, traffic forecasting, decision making for rescue people with disabilities in emergencies
...etc. [10], all these activities need working on huge data and a geographic data, to describe
human behavior and mobility patterns, for giving the simulation model more reality, since
almost simulation platforms are deal with only a grid to be the background of the model, so the
resulted model will be far away from reality.

30

Computer Science & Information Technology (CS & IT)

The results of this investigation show that the inter-event time t between each two successive
activities has bursty pattern, since there is long period without activities, this gives indication
about the population's heterogeneity. And human trajectories follows power-law distribution in
step size r, and they are modeled using displacements and waiting time distributions, as well as
the mobility traveling patterns are show that there are many short distances in contrast little long
distances. The mobile phone data is used to reconstruct the population trajectories. However, the
patterns of all days are very similar and they have approximately identical curves either in spatial
or temporal features. As next forward step with this kind of studies the mobility patterns would
be classified with regarding to the (work or off) days for more understandings to the life patterns.
Whereas, the radius of gyration would be considered as significant modelling parameter to give
the model more reality with focusing on patterns regularity.

REFERENCES
[1]

CS302 Lecture notes, (2015) Simulating Random Event,

http://web.eecs.utk.edu/~leparker/Courses/CS302-Fall06/Notes/PQueues/random_num_gen.html

[2]

Simulation: An Introduction, (2015) Stat. Chapter 10,

http://www.stat.nus.edu.sg/~zhangjt/teaching/ST2137/lecture/lec%2010.pdf

[3]

Sougata Biswas, (2014) Simulation of Exponential Distribution using R,https://rstudio-pubsstatic.s3.amazonaws.com/26693_e1151035722942b2813c0063c6b220ae.html

[4]
[5]

Vincent Zoonekynd, (2015) Probability Distributions, http://zoonek2.free.fr/UNIX/48_R/07.html

Solver, (2015) Simulation Tutorial Introduction, http://www.solver.com/simulation-tutorial

[6]

Siqi Shen, Niels Brouwers & Alexandru Iosup, (2011) Human Mobility in Virtual and Real Worlds:
Characterization, Modeling, and Implications, report number PDS-2011-007, ISSN 1387-2109.

[7]

The Pennsylvania State University, (2015) Probability Theory and Mathematical Statistics, STAT
414/415, https://onlinecourses.science.psu.edu/stat414/node/97.

[8]

Ilya Narsky, Frank C. Porter, (2013) Statistical Analysis Techniques in Particle Physics, Fits,Density
Estimation and Supervised Learning, Wiley-VCH, 1 edition.

[9]

Marta C. Gonzlez, Cesar A. Hidalgo & Albert-Laszlo Barabasi, (2009) Understanding individual
Human Mobility Patterns, nature 453.

[10] Gunther SAGL, Bernd RESCH, Bartosz HAWELKA, Euro BEINAT, (2012) From Social Sensor
Data to Collective Human Behaviour Patterns-Analysing and Visualising Spatio-Temporal Dynamics
in Urban Environments, GI_Forum 2012: Geovizualisation, Society and Learning. Herbert
Wichmann Verlag, VDE VERLAG GMBH, Berlin/Offenbach. ISBN 978-3-87907-521-8.