Preprocess

- Baseline correction

- Normalization

- BW ranking using Eq. (1)

Repeat p times

10-fold cross validation (CV)

• ECOC scheme

• Training

- Feature reduction using Eq. (7) and (10)

- Weight estimation using Eq. (6)

• Testing

- Decoding using Eq. (11)

- Find class label Ce which has the minimum distance using Eq. (12)

Repeat for each pair of classes

• PWC scheme

• Training

- Extended Markov blanket (EMB) feature selection using Eq. (16)

- Backward feature selection with features ranked by EMB

- Find the optimal feature subset S

- Increase the frequency of each feature that belongs to S by 1

• Testing

- Probability estimation using Eq. (14)

- Compute the final probabilities for all classes using Eq. (13)

- Find class label Cp which has the maximum probability using Eq. (15)

- Retrain with {Ce, Cp} samples if Ce <> Cp

Rank features according to the frequency for each pair of classes

Analysis of measurements