Random KNN

We present Random KNN, a novel generalization of traditional nearest-neighbor modeling. Random KNN consists of an ensemble of base k-nearest neighbor classifiers, each constructed from a random subset of the input variables. We study the properties of the proposed Random KNN. Using various datasets, we perform an empirical analysis of Random KNN performance and compare it with recently proposed methods for high-dimensional datasets. It is shown that Random KNN provides significant advantages in both the computational requirement and classification performance.