Information Technology Reference
In-Depth Information
Tabl e 6 . 22 A confusion matrix of a clustering solution obtained with LEGClust
for the DHN dataset.
Classes
176 169 3 0 0 1 3 0 1 9
1 90 0 0 0 000 0
7 1 188 1 16 2 178 14 3 6
0 3 3 195 83 2 0 0 5 0
1 0 3 4 101 10 4 1 7 0
02200 70311
0 1 0 0 0 2 0 60 1
0 2 1 0 0 8 0 53 1
0 0 0 0 0 8 14 0 180 1
530000110 1
The NCI Microarray dataset is a human tumor microarray data and an
example of a high-dimensional dataset. The data are a 64
6830 matrix of
real numbers, each representing an expression measurement for a gene (col-
umn) and a sample (row). There are 12 different tumor types, one with just 1
representative and three with 2 representatives. Experiments were performed
to compare the results from LEGClust with those obtained by Spectral clus-
tering. The final number of clusters for both algorithms was chosen to be 3,
following the example described in [94]. Results are shown in Table 6.21.
×
Tabl e 6 . 23 Results and parameters of LEGClust and Chameleon in experiments
on 4 real-world datasets.
Chameleon
LEGClust
an ARI
Mk
ARI
Iris
9
50
0.658
15
3
0.750
Olive
40
40
0.733
25
3
0.616
Wdbc
40
25
0.410
20
3
0.574
Wine
30
21
0.400
15
3
0.802
Results presented in Table 6.21 show that LEGClust performs better than
Spectral-Shi algorithm in the three datasets and, compared with Spectral-
Ng, it achieves better results in the DHN dataset and similar ones in the NCI
Microarray.
 
Search WWH ::




Custom Search