Mobility Data Mining - Mobility Data - page 125

need to be developed that exploit appropriate technologies, such as distributed
databases (a paradigm where data are distributed along several data centers, to
be queried to obtain the data needed for each specific analysis or computation
step), distributed computation (several nodes with computation powers collabo-
rate to analyze data), and streaming-oriented computation (essentially aimed to
perform computation by looking at the input data only once).
6.5 Bibliographic Notes
As mentioned at the beginning of the chapter, the literature on mobility data
mining is rather extensive - especially for such a young field - and heteroge-
neous. Attempting an exhaustive discussion of existing problems and proposals
would require much more space and would be beyond our purposes as well. In
the following, we will provide a list of essential bibliographic references for the
reader, including those describing the methods cited in the chapter and a few
pointers for further reading.
The original definition of flock patterns required that the group of objects meet
at a single time instant and have the same direction of movement. Successive
variants introduced the temporal duration constraint, also adopted in this chapter,
starting from Gudmundsson et al. ( 2004 ). Moving clusters were defined by
Kalnis et al. ( 2005 ), provided with a few heuristics for incrementally computing
the interesting patterns, while convoys are described in Jeung et al. ( 2008 )and
spatio-temporal sequential patterns appear in Cao et al. ( 2005 ).
T-patterns were introduced by Giannotti et al. ( 2007 ), and later were exploited
in buildingWhereNext - a location predictionmethod by Monreale et al. ( 2009 )-
as well as in several application works.
One rich source for a library of trajecory distances - to be used within generic
clustering algorithms - is provided by Pelekis et al. ( 2007 ). Several references
exist for standard (distance-based) clustering schema that can be applied to
trajectory data, including basic introductions to data mining such as Tan e t a l .
( 2005 ).
Model-based approaches to trajectory clustering can be found in several
isolated papers, especially on specific application domains (video surveillance,
animal tracking, etc.). The mixture-models trajectory clustering described in
this chapter was first introduced in Gaffney and Smyth ( 1999 ), later extended to
include time shifts. Hidden Markov models-based approaches can be found, for
instance, in Mlich and Chmelar ( 2008 ).
Time-focused clustering, an extension of density-based clustering for trajec-
tories, was presented in Nanni and Pedreschi ( 2006 ).
The TraClass framework for trajectory classification was introduced in Lee
et al. ( 2008a ), mainly based on previous works of the same authors on trajectory