Flexible Implementation of Multivariate Adaptive Regression Splines in Solving Continuous-State Stochastic Dynamic Programming

摘要

In this paper, the relaxation on the parameters of multivariate adaptive regression splines (MARS, [8]) is developed and implemented within the OA/MARS continuous-state stochastic dynamic programming (SDP) method [6]. The default stopping rule of MARS employs the maximum number of basis functions (M{sub}(max)), specified by the user. This paper presents two automatic stopping rules, which automatically determine an appropriate value for M{sub}(max), and a robust version of MARS that prefers lower-order interaction terms over higher-order terms. Computational results are presented on a wastewater treatment system application.