Algorithm 1:

Two-step feature selection of PredHS

Input:

Given the training set ,

Output:

Initialize to the subset of surviving features; ranked feature set

R = [ ];

Train a random forest with features in set S as input variables;

Calculate MDGI Z-Score for each feature in S;

Features with MDGI Z-Score smaller than 2.5 are removed from S;

whileS ≠ [ ] do

for each variable i in S, do

Train SVM classifiers on feature set S′, a subset of S exclusive of variable i;

Evaluate the ranking criterion Rc(i) of variable i;

end for

Rank the variable that maximizes Rc;

best = arg maxiRc;

R = [best R];

Remove the variable best from the set S;

end

Return the ranked feature set R.