I have a question:

If you get the top 500 significant genes applying SVM-rfe (or any other algorithm). Those top 500 genes are ranked as how many times they appear on top for a number of iteration (let say 100 iterations). The results are :

Sensitivity  : 0.82

Specificity :0.69 Accuracy : 0.77

Now if consider just those top 500 genes and run the same procedure to pick the top 300 (or less or more) then I get those results:

Sensitivity  : 0.919

Specificity :0.921 Accuracy : 0.920

How do you explain this?