Fast knot optimization for multivariate adaptive regression splines using hill climbing methods

Xinglong Ju, Victoria C. P. Chen, Jay M. Rosenberger, Feng Liu

Abstract: Highlights•Two novel methods, PHCM and HCM, for MARS knot positioning are proposed.•Traditional and two novel methods are similarly accurate and robust with noise.•The proposed methods are 10% to 70% faster than traditional method.•PHCM is 70% faster than HCM for higher dimensional datasets.•The Python code will be made open-source on GitHub to help other researchers.