Basedonanexactintegralexpressionfortherisk,anasymptoticevaluationoftheconditionalriskisderivedfordistributions with have unbounded supports, which, using Laplace’s method. Then, by integrating these asymptotic expansions, we evaluate the asymptotic evaluation of the finite sample risk (the unconditional probability error). The finite sample risk for the Pareto and exponential distributions are discussed as typical.
Digital Object Identifier (DOI)
M. Rizk, Mohamed
"On the Asymptotic Evaluation of the Finite-Sample Risk of the Nearest Neighbor Classifier,"
Applied Mathematics & Information Sciences: Vol. 16:
4, Article 12.
Available at: https://digitalcommons.aaru.edu.jo/amis/vol16/iss4/12