Table 2.
Classifier family | Parameters | Ranges | Step |
---|---|---|---|
Naive Bayes | – | – | – |
k-NN | k | [0; 100] | linear |
J48 | Ninstances, Cpruning | [0; 5], [0.02; 0.24] | lin, lin |
MLP | Rlearn, Nhidden | [10−10 ; 105], {0, 6, 22, 45} | log, – |
SVM | C, σ | [1; 100], [102; 10−2] | lin, log |
The values for the number of hidden layer units, Nhidden, of the MLP are chosen as {none, number of classes, (number of attributes + number of classes)/2, number of attributes + number of classes}