A new method for feature selection based on weighted [formula omitted]-nearest neighborhood rough set

Abstract

The neighborhood rough set theory is a helpful instrument for working with data that is numerical, and the performance of its uncertainty measures is generally stable. Even one noisy sample may result in several samples into the border domain because of the partitioning approach for upper approximation and lower approximation in neighborhood rough sets, which is particularly sensitive to incorrectly labeled samples. Based on the distributional information about the data in the sample area, the Weighted k-nearest Neighborhood Rough Set (WKNRS) model was proposed as a solution to this problem. Firstly, we weight the neighbor samples using the standard deviation of pertaining class labels before offering a sample quality evaluation. Next, using this sample quality evaluation, a WKNRS is proposed, which fixes the neighborhood rough set’s noise sensitivity issue. Then, a Weighted k-nearest Neighbor Features Selection(WKNFS) is generated using this model, along with a forward greedy search methodology. Finally, because we wanted to assess this method’s efficiency, we put it to the test using 16 UCI datasets. The results of the trials demonstrate that this algorithm performs better than the other 6 algorithms of feature selection currently in use.