FRL: An Integrative Feature Selection Algorithm Based on the Fisher Score, Recursive Feature Elimination, and Logistic Regression to Identify Potential Genomic Biomarkers
Procedure 1
Procedures of FRL (mainly illustrated bymatrix).
Method: FRL
Input: Gene expression original matrix
Output: Gene subset
For in 1 : 10
Pretreat and Apply Robust Multi-Array average to remove the effect of labeled probes on real gene expression and form matrix
End
For in
Calculate the fisher score of based on formula Equation (2)
End
Divide the same maximum number features as one cluster. Spot the inflection point and extract the abundant data. Name new matrix as
Set up the iteration step and according to diverse dimension
For in
For in -fold
Train recursive feature elimination model with logistic regression classifier in
Calculate accuracy in
Use -fold cross validation to calculate average
End
End
Obtain the current dimension with the optimal performance
Select the feature subset according to to make matrix
Extract the repeated genes from and make a new matrix