Determination of Support Vector Boundaries in Generalized Maximum Entropy: An Alternative Approach

Abstract

The most important question in the Generalized Maximum Entropy approach is how to determine the support vector boundaries. The purpose of this study is the determination of these boundaries in order to help the researchers to obtain more consistent parameters than classical methods, without any prior information about the parameters. This study suggests that with the use of binomial and normally distributed datasets, a new procedure can be used to determine the support vector boundaries for datasets each having different characteristics. As a result, the suggested approach provides more consistent parameter estimations than the classical methods in both small and large samples with and without the existence of outliers.