Stability analysis of WkNN feature selection

Recently, huge amounts of data have been generated by computer and internet applications in multiple domains, including healthcare, bioinformatics, social media, e-commerce, and transportation. These data often have characteristics of high dimensions and their analysis is a challenge for researchers in the fields of machine learning and data mining. Feature selection is a dimensionality reduction technique that aims to select a subset of relevant features from the original feature set. Important aspect of feature selection, besides ability to identify all significant features, is the stability of the feature selection. In this paper, we investigate the stability of the Weighted k-Nearest Neighbors feature selection and compare it to other state-of-the art feature selection methods.