A Density-Based Clustering of Spatio-Temporal Data - New Contributions in Information Systems and Technologies - page 42

Figure 4 represents the output clusters resulting from carrying out the optimum run
for the hurricane data set. This Figure represents clusters formed from the run with
eps=40 and MinLns=8, which gives the least Quality Measure (optimum). In Figure 4
below, there are 6 representative lines, one representing each cluster.
Until now, only the spatial parameter of each segment is analyzed. The next target
is to analyze the time dimension as well. The segments that fall in each cluster are
printed in an Excel sheet to be able to analyze each cluster separately and make
predictions not only about their location, but also their time of arrival. The start and
end points of each line segment in each cluster is printed. The maximum start time,
minimum start time, range, average, variance, and the standard deviation of all line
segments in each cluster are calculated for the time dimension.

Fig. 4. Clustering results for Hurricane Data

From Table 1, a couple of conclusions are reached. Cluster 1, even though the largest
in size and has the highest density of line segments within it, has an average segment
time of travel of 2.84 days. That means, it is most likely that a hurricane passing in the
direction of Cluster 1 will take the same path of the line representing Cluster 1 and will
arrive in 2.84 days approximately from the beginning time of its motion. It is also
noticed that this cluster is spatially farther away from all other clusters. Cluster 2, has 20
segments in it and the greatest average segment time of travel, 4.05 days.

Table 1. Summary of the Temporal Dimension for Hurricane Data

For eps=40
& MinLns=8

Number of
Segments

Average Segment Time

Cluster 1 889 2.84 days

Cluster 2 20 4.05 days

Cluster 3 15 2.78 days

Cluster 4 22 3.51 days

Cluster 5 21 3.46 days

Cluster 6 16 3.11 days