Improving fuzzy-rough quick reduct for feature selection

Feature selection is a process of selecting subset of features which are highly correlated with classification outcome and lowly depends on other features. Rough set has been successfully applied to nominal datasets for feature selection. Since datasets might have real-valued data, Fuzzy set theory has been combined with Rough set for feature selection of continuous datasets. Fuzzy Rough Set Feature Selection (FRFS) is computationally prohibitive. Many researchers proposed new methods to diminish the computation of FRFS. A new method based on Fuzzy Lower Approximation-Based Feature Selection is proposed which selects smaller subset of features, makes better classification accuracy and run faster than the base method, especially on big datasets. This is performed using a threshold based stopping criterion which prevents adding more features in QuickReduct algorithm. Experimental results on UCI datasets confirm the performance and effectiveness of our proposed method.