R. Blanquero, E. Carrizosa, P. Ramirez Cobo, R. Sillero-Denamiel
In this work we delineate novel applications of Bayesian inference techniques in the context of statistical classification. Specifically, we care about classification problems where imbalancedness is present, at either the classes size or the misclassification cost structure. On one hand, we explore estimation of performance measures of relevance (the misclassification rates, predictive values) in Bayesian way, which seems for adequate for accommodating the cost structure and the eventual imbalancedness of the data set than the traditional frequentist estimates .On the other hand, we present a novel formulation of the Naïve Bayes classifier in terms of an optimization problem where new constraints (possibly nonlinear and based on integer variables) for controlling the performance measures are added. Finally, applications for real metabolomics data sets will be discussed.
Palabras clave: Naive Bayes, Costs, Imbalancedness, Mixed Integer Nonlinear Programming
Programado
L06.2 Grupo de Inferencia Bayesiana: Bayesian model selection in Classification and Clustering
5 de septiembre de 2016 12:55
0.09 - Aula de proyectos 2
