Table 1 The leave-one-out and 5-fold cross validation classification accuracies of the SVM-classifier and the KNN-classifier based on four gene selection methods, GS1, GS2, Cho's, and F-test, on the SRBCT dataset. Listed are the accuracies when the numbers of selected genes are 30, 60, and 100, respectively, together with their standard deviations for 5-fold cross validation, and the best accuracy together with the number of selected genes.
SRBCT 5-Fold | KNN | SVMs | ||||||
|---|---|---|---|---|---|---|---|---|
| Â | 30 | 60 | 100 | Best Accuracy/# Genes | 30 | 60 | 100 | Best Accuracy/# Genes |
GS2 | 0.953 ± 0.048 | 0.971 ± 0.041 | 0.980 ± 0.038 | 0.981/90 | 0.949 ± 0.047 | 0.976 ± 0.040 | 0.990 ± 0.026 | 0.990/99 |
GS1 | 0.941 ± 0.047 | 0.961 ± 0.045 | 0.977 ± 0.041 | 0.980/88 | 0.959 ± 0.054 | 0.978 ± 0.040 | 0.988 ± 0.030 | 0.979/93 |
Cho's | 0.820 ± 0.096 | 0.864 ± 0.093 | 0.896 ± 0.087 | 0.902/98 | 0.835 ± 0.088 | 0.918 ± 0.069 | 0.943 ± 0.062 | 0.943/98 |
F-test | 0.963 ± 0.050 | 0.973 ± 0.046 | 0.978 ± 0.040 | 0.980/90 | 0.970 ± 0.042 | 0.980 ± 0.039 | 0.992 ± 0.021 | 0.992/95 |
SRBCT LOO | KNN | SVMs | ||||||
| Â | 30 | 60 | 100 | Best Accuracy/#Genes | 30 | 60 | 100 | Best Accuracy/# Genes |
GS2 | 0.964 | 0.976 | 0.964 | 0.988/77 | 0.952 | 0.976 | 1.000 | 1.000/96 |
GS1 | 0.964 | 0.988 | 0.988 | 0.988/57 | 0.976 | 0.988 | 0.988 | 0.988/34 |
Cho's | 0.831 | 0.880 | 0.892 | 0.928/82 | 0.819 | 0.928 | 0.964 | 0.988/80 |
F-test | 0.976 | 0.976 | 0.988 | 0.988/89 | 0.976 | 0.980 | 0.988 | 1.000/78 |