Weighted k-nearest neighbors feature selection for high-dimensional multi-class data

Feature selection is considered as one of the important steps in processing of high-dimensional data. Identification of significant genes in micro-array sequences, and dimensionality reduction in multimedia data are the example areas where the feature selection is beneficial. In this paper, we present set of methods based on distance and attribute weighted k-nearest neighbors algorithm. The new methods are obtained by deployment of additional distance measures and loss functions. Moreover, we present some extensions of original weighted k-nearest neighbors feature selection method such as multi-class classification, regularization, and computationally effective TensorFlow implementation.