Automatic Nearest Neighbor Rules
The error probability of the k-nearest neighbor rule converges to the Bayes risk for all distributions when k → ∞, and k/n → 0 as n → ∞. The convergence result is extended here to include data-dependent choices of k. We also look at the data-based selection of a metric and of weights in weighted nearest neighbor rules.
KeywordsError Probability Training Sequence Positive Definite Matrix Neighbor Rule Empirical Error
Unable to display preview. Download preview PDF.