COMPSTAT 2016: Start Registration
View Submission - CRoNoS FDA 2016
A0157
Title: Bayesian functional linear regression with sparse step functions Authors:  Paul-Marie Grollemund - University of Montpellier (France) [presenting]
Christophe Abraham - Montpellier SupAgro (France)
Pierre Pudlo - University of Aix-Marseille (France)
Meili Baragatti - Montpellier SupAgro (France)
Abstract: Scalar-on-function regression is a common tool to explain scalar outcomes from functional predictors. It is helpful in many applications, for instance in agronomy to explain the yield of parcels from temperature curves. Our aim is to estimate the coefficient function with an interpretable estimate. The idea is to recover the relevant intervals of the support that explain the scalar outcome. For instance it is important for farmers to know the periods during which temperature plays a main role in the yield of their parcels. We will first define the notion of an interpretable estimate and present the Bayesian model. Then we will present the estimator and the numerical process to select intervals. We will also propose a way to take prior knowledge into account. Eventually the proposed method will be applied on simulated datasets and real world datasets.