Database Reference
In-Depth Information
(a)
(b)
Figure 10.6 The result of clustering from the trajectories moving from the center to the
northeast area. (a) The input data set for the clustering algorithm: the trajectories moving
from the center to the northeast area. (b) The resulting clusters using the
Route Similarity
distance function. The clusters are visualized using a themed color, and the analyst can
select and browse them separately.
it might arise only on Monday) or might have a general validity. In order to dis-
tinguish these two cases, we need to measure how the population of the clusters
is distributed over the days of the week, and this task can be accomplished using
the clustering as an unsupervised classification model. More precisely, after the
clusters have been extracted for a specific day, one or more representatives,
named specimens, for each cluster are computed and such representatives are
used to classify the trajectories in other days of the week: every new (unseen)
trajectory
T
is classified by assigning it to the closest specimen (and therefore
to the cluster it represents). If the distance between
T
and such a specimen is
too high, however, the trajectory is assigned to the
noise
. Figure
10.7
shows how
