Nonparametric Classification for Circular Data

Local methods of classification are of interest when a simple binary rule does not fit data or basic parametric assumptions are not satisfied. Despite its practical relevance, this topic has still not been explored in the directional context. We discuss some nonpara- metric approaches to classify circular data. They rely on kernel estimation of circular densities and local regression with binary response and circular predictor. We provide some asymptotic L2 properties for the proposed methods along with simulation results and a real-data application.