Model Design. This figure describes the methodology used in the testing. The methodology employs a similar wrapper technique as described in . The data sets were first divided, in a stratified way, into training and hidden test data (80% training and 20% test, except for Alon which was 75%/25% since it was smaller). The weight vector from the SVM with a linear kernel was used to identify the gene(s) to remove. The training data was projected down to just the specified subset and a classifier was constructed using a linear SVM. Then the test data was projected to the same feature set and tested using the classifier built from the training data. This was repeated for each data set, for each algorithm and for each size feature set from 1 to the maximum features selected.