New approaches to nonparametric circular regression models

Abstract

Nonparametric regression models are employed to examine the dependence between two or more random variables, without assuming a specific form for the regression function. However, complex data structures often arise in practice, leading to situations where the support of the variables is not Euclidean. This is the case of circular variables, defined on the unit circumference. Classical nonparametric regression methods do not take into account the periodicity of the data, and thus are not adequate for this kind of observations. This thesis provides new nonparametric regression models and inference tools to deal with circular variables. The performance of the proposed methodologies is analyzed and illustrated with real data applications.