Automatic knot selection in smooth additive models

N. Carrizosa, M. Durbán, V. Guerrero

B-Spline regression is one of the most powerful and flexible methods in nonlinear regression. This methodology requires deciding the number and placement of changepoints - known as knots - before the estimation process. Traditionally, this has been done manually or addressed using a penalty term to control the regressor’s smoothness (P-Splines). In this work, the algorithm for automatic knot selection (A-splines) proposed in Goepp et al. (2025) for univariate regression with a continuous response variable is extended to the case of generalized additive models. Our approach is successfully tested in both synthetic and real datasets.

Keywords: B-splines, Generalized additive models, Knot selection

AMC4 Prediction and Classification
June 11, 2025  10:30 AM
MR 1

L. Vicente Gonzalez, E. Frutos Bernal, J. L. Vicente Villardón

F. Scielzo Ortiz, A. Grané, I. Albarrán

A. Jiménez-Cordero, S. Pineda, J. M. Morales González

M. D. Jiménez Gamero, M. R. Sillero Denamiel