Binary weighted mean of vectors optimization based type-2 fuzzy-rough for feature selection

One of the crucial problems in in the fields of machine learning and data mining is data reduction by feature selection (FS). In this context, this paper proposes an FS method based on a hybrid of type 2 fuzzy rough k-nearest neighbors (T2FRKNN) and a weighted mean vector optimization method called FKNINFO. Thus, the significance of the features can be determined by the creation of the lower and upper fuzzy similarity partition matrices. The introduction of INFO is intended to enhance the T2FRKNN with the best parameters and feature subsets. The proposed method is a dynamic framework originally aimed at solving problems through continuous optimization. In this regard, we propose a binary version of FKNINFO (BFKNINFO), which uses the X-shaped function to improve the efficiency of FS. The BFKNINFO is tested using medical datasets and compared to the other optimization methods in terms of fitness, accuracy, precision, recall, ROC curves,Wilcoxon statistical test (P-value), running time, and number of features. BFKNINFO is used to detect the coronavirus disease (COVID-19) datasets. The results of the experiments demonstrate the effectiveness of BFKNINFO in navigating the problem space and identifying the most effective parameter and features by reducing the number of features.