Modeling with MARS and CMARS

Fatma Yerlikaya-Ozkurt

Institute of Applied Math, Scientific Computing Program

May 24, 2012, Thursday, 14:00-15:30
Institute of Applied Math, S209

MARS: Multivariate Adaptive Regression Splines
CMARS: Conic (Convex, Continuous) Multivariate Adaptive Regression Splines


Conic (Convex, Continuous) Multivariate Adaptive Regression Splines (CMARS), developed at Institute of Applied Mathematics (IAM), Middle East Technical University (METU), is an alternative approach to the Multivariate Adaptive Regression Splines (MARS). MARS is a well-known data mining tool capable of modeling high-dimensional data with nonlinear structure. Flexible nature of MARS modeling leads successful implementation of the method in various application areas. On the other hand, CMARS is based on a penalized residual sum of squares (PRSS) for MARS as a Tikhonov regularization (TR) problem. CMARS treats this problem by a continuous optimization technique, in particular, the framework of Conic Quadratic Programming (CQP). These convex optimization problems are very well-structured, herewith resembling linear programs and, hence, permitting the use of interior point methods. CMARS and MARS methods are preferred by modelers and data analysts in the fields of nonparametric regression/classification, and data mining. By this talk, the implementation of MARS and CMARS methods will be provided on test data. For CMARS algorithm, a user-friendly computing environment which is written in MATLAB will be used. For MARS, R program, a well-known and popular statistical software, will be preferred.