46 /342
為頻繁單變量不確定樣式產生摘要
46
When the two clusters are merged, the SFC algorithm keeps merging clusters by
iterating the above process. If one or two of the criteria are not met, the merging process is
stopped.
3.3.5 Step 6: Summary of the FU2Ps
The final step is generation of the summary. To return an informative summary of all
FU2Ps to users, we retrieve the following data from each cluster: 1) the representative FU2P
(medoid); 2) the lower bound and upper bound of the expected supports of the FU2Ps in the
cluster; 3) the maximal range of each attribute's interval in the cluster; and 4) the maximal
number of attributes in a FU2P in the cluster. We use an example to illustrate the proposed
summary. Table 8(a) presents the clusters derived by applying the SFC algorithm to the set of
FU2Ps presented in Table 6 where
w
,
ξ
, and
δ
are set as 0.1, 0.25, and 1.1, respectively. In
Figure 8(a), the column “Medoid” denotes the medoid of each cluster; the column “Cluster
members” denotes the FU2Ps in the clusters, each of which is denoted by its expected
support shown in a pair of parentheses. Please note that the medoid of each cluster is also a
member of the cluster. Because
w
is set as 0.1, most FU2Ps in a cluster are similar to each
other in appearance. However, some FU2Ps are assigned mainly according to their expected
supports because they are not similar to any medoid in appearance, such as [
A
2
:[335, 337]] in
the second cluster. The summary of the first cluster comprises: 1) the representative U2
pattern ([
A
4
:[361, 367]]); 2) the lower bound (1682.998) and upper bound (3613.998) of the
expected supports; 3) the maximal range of intervals in attribute
A
2
, and
A
4
([335, 337] and
[361, 374], respectively); and 4) the maximal number of attributes in a FU2P (2). The
maximal range of an attribute's interval in the cluster comprises the minimal lower bound
and maximal upper bound of the attribute's intervals appearing in the cluster, which describes
the extent of the interval on this attribute. That is, the FU2Ps in this cluster are formed by the
attributes with intervals within the respective maximal ranges of these attributes. In addition,
each of the FU2Ps in the cluster at most comprises two attributes. The expected supports of
the FU2Ps in this cluster are between the lower bound and upper bound of the expected
supports. The summaries derived from the five clusters construct the overall summary.