Database Reference
In-Depth Information
4. How can we detect an extraordinary event and understand the associated
mobility behavior? How and when do people reach and leave the event's
location? What is the spatio-temporal distribution of such (portion of) trips?
5. What will be the areas with highest traffic volume in the next hour(s)? To
what extent are our predictions accurate?
6. Are there geographic borders that emerge from the way people use the terri-
tory for their daily activities? If so, how do we define such borders? Do these
borders match the administrative ones?
More than just examples, these questions are paradigmatic representatives of
the analysts' need to disentangle the huge diversity of individual whereabouts
and discover the subgroups of travels characterized by some common behavior
or purpose. It is no surprise, then, that finding answers to these questions is
beyond the limits of the current generation of commercial systems, and cannot
even be accomplished by simply applying single known research prototypes,
such as the mobility data mining methods presented in Chapter
6
. There is the
need for a
mobility knowledge discovery process
aimed at
discovering interest-
ing subgroups of vehicles and travels characterized by some common movement
behavior
. To perform this kind of analysis, a complete querying, analysis, and
mining system is needed, able to support the overall knowledge discovery pro-
cess centered around the trajectory concept. In this chapter we will provide ana-
lytical answers based on the tools and the knowledge discovery process handled
by an analytical framework named M-Atlas, already introduced in Chapter
7
.A
general analytical process on mobility data follows several steps. First the data
are explored by the analyst to understand and comprehend the several dimen-
sions of the observed phenomena. In Section
10.3
we present a set of statistical
methods that have a twofold objective: on one hand they serve to assess the
general validity of the data with respect to background knowledge; on the other
hand they provide insight into the internal distribution of data dimensions. Once
the analyst has acquired a deep understanding of the data, he or she can proceed
with the exploration. Section
10.4
provides a set of analytical scenarios where
different mobility data mining methods are used to find answers to the questions
we have proposed. The methods used have already been presented in Chapter
6
,
thus we refer the reader to that chapter and we will not give further details on
the internal functionalities of each algorithm.
To present a paradigmatic mobility knowledge discovery process we concen-
trate on massive, real-life GPS data sets, obtained from tens of thousands of
private vehicles with on-board GPS receivers. The owners of these cars are sub-
scribers to a pay-as-you-drive car insurance contract, under which the tracked
trajectories of each vehicle are periodically sent (through the GSM network)
to a central server for antifraud and antitheft purposes. This data set has been
