Database Reference
In-Depth Information
applications of the trajectory data mining methods we describe here in the next
chapters, especially Chapters
7
,
9
,and
10
.
6.2 Local Trajectory Patterns/Behaviors
The mobility data mining literature offers several examples of trajectory patterns
that can be discovered from trajectory data. Among this wide variety, a very
large number of proposals actually adopt two basic assumptions: first, a pattern
is interesting (and therefore extracted) only if it is frequent, and therefore it
involves (or appears in) several trajectories
1
; second, a pattern must describe
(also) the movement in space of the objects involved, and not only aspatial or
highly abstracted spatial features. In this chapter we will adopt such assumptions,
in order to better focus the discussion.
While the spatial component of trajectory data is typically part of the patterns
extracted, the temporal one (also intrinsic in trajectory data) can be treated in
several different ways, and we will use this differentiation to better organize the
presentation. Then, while a trajectory pattern always describes a behavior that is
followed by several moving objects, we can choose whether they should do so
together (i.e., during the period), at different moments yet with the same timing
(i.e., there can be a time shift between the moving objects), or in any way, with
no constraints on time.
6.2.1 Using Absolute Time or Groups That Move Together
One of the basic questions that arise when analyzing moving objects trajectories
is the following:
Are there groups of objects that move together for some time?
For instance, in the realm of animal monitoring such kind of patterns would
help to identify possible aggregations, such as herds or simple families, as well
as predator-prey relations. In human mobility, similar patterns might indicate
groups of people moving together on purpose or forced by external factors, for
example, a traffic jam, where cars are forced to stay close to each other for a
long time period.
Obviously, the larger the groups and/or the longer the period they stay
together, the higher the likelihood that the observed phenomenon is not a pure
coincidence. For instance, if two members of a population of zebras under mon-
itoring happen to move close to each other for a short time, that can be seen
as a random encounter. However, if dozens of zebras are observed together for
1
Of course, significant exceptions exist, including the extreme case of outlier detection, consisting of
anomalous (and thus infrequent) patterns. For ease of presentation, outlier detection will be described
later in this chapter, in the context of
global models
.
