The Research on an Adaptive k-Nearest Neighbors Classifier

K-nearest neighbor (KNNC) classifier is the most popular non-parametric classifier. But it requires much classification time to search k nearest neighbors of an unlabelled object point, which badly affects its efficiency and performance. In this paper, an adaptive k-nearest neighbors classifier (AKNNC) is proposed. The algorithm can find k nearest neighbors of the unlabelled point in a small hypersphere in order to improve the efficiencies and classify the point. The hypersphere´s size can be automatically determined. It requires a quite moderate preprocessing effort, and the cost to classify an unlabelled point is O(ad) + O(k)(l les a Lt N). Our experiment shows the algorithm performance is superior to other known algorithms