Information Technology Reference
In-Depth Information
4.6 Apply filters to E
ci
5
Filter using FF with clustering settings N=5 clusters, seed 1. The current Diff=45%>d.
Change to Expectation Maximization (EM) as the next filter.
4.7 Apply filters to E
ci
6
Filter using EM. The clustering settings are min stdDev :1.0E-6, num clusters: -1 (automatic),
seed: 100. The current Diff=13%>d, Log likelihood: -18.04546. Split E
ce
and filter the more
cohesive
8
subset of E
ci
s using EM. Log likelihood: -17.99898, diff=8.
Cluster
E
ci
1
E
ci
2
E
ci
3
E
ci
4
E
ci
5
E
ci
6
E
ci
6
*
0
1 ( 17%)
3 ( 33%)
3 ( 20%)
3 ( 20%)
2 ( 6%)
13 ( 29%)
8 ( 18%)
1
1 ( 17%)
1 ( 11%)
5 ( 33%)
5 ( 33%)
4 ( 11%)
7 ( 16%)
8 ( 18%)
2
1 ( 17%)
2 ( 22%)
2 ( 13%)
2 ( 13%)
9 ( 26%)
11 ( 24%)
12 ( 26%)
3
2 ( 33%)
2 ( 22%)
3 ( 20%)
3 ( 20%)
18 ( 51%)
14 ( 31%)
8 ( 18%)
4
1 ( 17%)
1 ( 11%)
2 ( 13%)
2 ( 13%)
2 ( 6%)
9 ( 20%)
diff
16,00%
11,00%
10,00%
20,00%
45,00%
13,00%
8,00%
*This is the result of EM to define the splitting of E
ce
1.
Table 4. Filtering results for each E
ci
4.8 Build E
ce
2 to E
ci
6
Keep all the individuals as E
ci
1, and put in E
ce
2 the individuals in cluster 1 (one of the three
less cohesive, with lower p
o
). This procedure is shown in Figure 6.
Fig. 6. E
ce
1 after cleaning up the less cohesive E
ci
s
8
Cohesiveness is defined according to MLW as distance and sequence of filters. In this case it is
implemented using EM forcing 5 clusters, and selecting the four clusters with more elements.