Information Technology Reference
In-Depth Information
(a)
0
100
200
300
400
500
600
700
0
100
200
300
400
500
600
700
Time
(b)
0
100
200
300
400
500
600
700
0
100
200
300
400
500
600
700
0
100
200
300
400
500
600
700
0
100
200
300
400
500
600
700
Time
Fig. 12.7
Medoid time series of biggest cluster (with
k
=
2) found by our RRR distance measure
for
a
univariate and
b
multivariate case. The intervals highlighted in
red color
indicate patterns that
frequently recur in the time series objects of the corresponding cluster, whereas intervals in
blue
indicate low recurrence
of clusters, which corresponds to determinism value of 0
16, respectively.
As might be expected, the results for the univariate time series are better than for
the multivariate case, because the search space expands and the probability of recur-
ring patterns decreases with an increasing number of dimensions or measurements,
respectively. In both cases, however, our RRR distance performs about 10% better
than the compared DTW distance, meaning that the identified prototypes contain
10% more recurring (driving behavior) patterns.
Figure
12.7
shows the prototype or rather medoid time series of the biggest cluster
found by the k-medoids algorithm (for
k
.
26 and 0
.
=
2) in combination with our RRR distance
Search WWH ::
Custom Search