Table 3.
Accuracy of a dozen of different combinations of feature selection and learning methods
| Feature Selection Methods | |||||
|---|---|---|---|---|---|
| Information Gain | MeanDiff | mRMR | PCA | ||
| Learning Methods | Decision Tree | 50.88% | 52.06% | 51.20% | 51.69% |
| KNN | 56.17% | 58.71% | 57.78% | 51.36% | |
| SVM-RBF | 55.37% | 57.30% | 56.18% | 51.84% | |
10-fold cross validation accuracies of combination of 4 feature selection methods and 3 learning methods shows that none of these combinations are more accurate than our suggested combination of MeanDiff500 feature selection and BestKNN learning (59.55%); indeed, several do not even beat the baseline of 51.52%.