A Novel Hybrid Dimension Reduction Technique for Undersized High Dimensional Gene Expression Data Sets Using Information Complexity Criterion for Cancer Classification
Table 2
Classification results of benchmark gene expression data sets using the 5, 10, and 15 original genes.
Data sets
Original number of dimensions
LDA misclassification error rates
QDA misclassification error rates
5 genes
10 genes
15 genes
5 genes
10 genes
15 genes
Leukemia
3571
43.1%
33.3%
20.1%
31.9%
22.2%
6.9%
Colon
2000
35.4%
32.2%
25.8%
32.2%
16.1%
16.1%
Prostate
6033
34.3%
36.27%
18.6%
30.4%
25.5%
18.6%
Lymphoma
4026
30.6%
29.0%
16.1%
19.4%
1.61%*
3.22%*
SRBCT
2308
31.8%
12.7%
17.5%
15.9%
3.18%*
0.0%*
Brain
5597
45.2%
30.9%
14.2%
42.8%*
9.5%*
16.6%*
There are singularities in the covariance matrices.
