23º SINAPE - Simpósio Nacional de Probabilidade e Estatística

Página Inicial » Inscrições Científicas » Trabalhos

Dados do Trabalho


Título

A U-STATITICS BASED TEST FOR ONE-CLASS CLASSIFICATION IN HIGH DIMENSIONAL DATA

Resumo

We consider the issue of statistical testing in classification problems for high dimensional low sample size data, for which a U-statistics based approach has shown promise. Here, we address One Class Classification (OCC) problems, which consist of verifying whether an object represented by a vector of feature values can be assigned to a target class. To this objective, we consider a U-statistic defined on within and between group distances, and present an extension to consider groups of size one. We derive the asymptotic properties for this extension, and employ it to define a model free classification test. We also present a simulation study to asses the test's statistical properties and perform an application to real data.

Palavras-chave

U-Statistics, Asymptotic Theory, Outliers

Área

Inferência Estatística

Autores

Marcio Valk, Gabriela Bettella Cybis