Mobile Network Data Analytics for Intelligent Transportation Systems

In this dissertation, we explore how the interplay between transportation and mobile
networks manifests itself in mobile network billing and signaling data, and we show how
to use this data to estimate different transportation supply and demand models.
To perform the necessary simulation studies for this dissertation, we present a simula-
tion scenario of Luxembourg, which allows the simulation of vehicular Long-Term Evolu-
tion (LTE) connectivity with realistic mobility.
We first focus on modeling travel time from Cell Dwell Time (CDT), and show –
on a synthetic data set– that we can achieve a prediction Mean Absolute Percentage
Error (MAPE) below 12%. We also encounter proportionality between the square of
the mean CDT and the number of handovers in the system, which we confirmed in the
aforementioned simulation scenario. This motivated our later studies of traffic state models
generated from mobile network data.
We also consider mobile network data for supporting synthetic population generation
and demand estimation. In a study on Call Detail Records (CDR) data from Senegal,
we estimate CDT distributions to allow generating the duration of user activities, and
validate them at a large scale against a data set from China. In a different study, we
show how mobile network signaling data can be used for initializing the seed Origin-
Destination (O-D) matrix in demand estimation schemes, and show that it increases the
rate of convergence.
Finally, we address the traffic state estimation problem, by showing how handovers can
be used as a proxy metric for flows in the underlying urban road network. Using a traffic
flow theory model, we show that clusters of mobile network cells behave characteristically,
and with this model we reach a MAPE of 11.1% with respect to floating-car data as ground
truth. The presented model can be used in regions without traffic counting infrastructure,
or complement existing traffic state estimation systems.