Using Feature Selection Based on Mutual Information, Max-Relevance, Min-Redundancy MRMR returns features which is ranked. How can I select a subset of these features that maximize the performance of classifiers based on threshold?
i.e - ranked feature f1, f2 and f3 give 90% Accuracy .
ranked feature f1, f2,f3 and f4 give 88% Accuracy .
ranked feature f1, f2 , f3, f4 and f5 give 92% Accuracy .