Boosting Supervised Learning Problems in the High Dimensional Context
09/05/2022, 11:00, K71
We propose a novel domain selection methodology for high-dimensional and functional data. We introduce the divergence curve as a tool to drop out redundant information in the context of supervised classification problems. The proposed method learns and infers about the sub–interval of the domain that better discriminates the classes of functions. Simulations results show that the proposed methodology improves the classification performance reducing, at the same time, the computational burden of several functional classification methods. We apply the methodology to several data sets and the empirical results show remarkable improvements in supervised classification when the effective domain is learned as a first step of the process.