Algorithm CPEMM (DS, threshold) |
Input: DS → dataset, merit_threshold meritvalue, |
acc_threshold accuracy threshold, flag = 1 |
Output: ffs finalised feature set |
Accuracy vector, MS merged set vector, |
bs bootstrap vector |
merit subsets obtained from wrapper feature selection, |
Sorted subset list in descending order |
subset with highest ranking, |
no of subsets generated () |
(1) Initialize subset = null |
(2) Generate feature subsets with Ensemble |
Initialise no. of iterations |
Repeat |
Generate subset from DS |
Verify merit |
|
Optimization of feature subset |
merit of |
Until iterations |
(3) Evaluate accuracy of classifier with subset . |
, , |
Evaluate accuracy of |
Repeat |
Repeat |
diff = |
and merged |
Evaluate accuracy of |
If |
Append to ffs |
flag = 1 |
else |
flag = 0 |
Append to ffs |
endif |
Increment |
until flag != 0 or or diff threshold |
Increment |
until ≥ acc_threshold |
(4) bootstrap vector from ffs |
Repeat |
Train classifier ci with |
Evaluate out of bag error |
Until sets are bootstrapped |
with 5-fold cross validation |