Improving k-nearest neighbor Bayes error estimates

Studies the k-nearest neighbor density estimation with fixed k, and its use in Bayes error estimation as proposed by Fukunaga and Hummels (1987). The authors arrive at a modified formula for a density estimator by analyzing the bias and variance expressions. The authors use four modifications of this formula for Bayes error estimation, and perform controlled experiments on Gaussian data in order to compare its performance with the original method. Given experimental results justify the proposed modifications.