Appendix B. Information about multivariate adaptive regression splines (MARS) and methods used in model development.
In MARS, different splines (line segments representing different regression relations within a single basis function) are joined at locations called knots, which are values of explanatory variables where the slope of the basis function changes. For each basis function, knots for splines are identified through an intensive automated search. To account for this intensive search, MARS computes effective degrees of freedom for each basis function. To determine the effective degrees of freedom that should be charged to each basis function, we applied the 10-fold cross validation procedure available in MARS2.0, an approach that protects against overfitting and against fitting a model that omits important variables (Salford Systems 2001). This cross validation procedure in MARS is designed to determine the appropriate number of degrees of freedom for each basis function and was not used to test the predictive ability of the model; the five Test data sets were used to assess the model’s predictive ability.
Salford Systems. 2001. MARS user guide. Salford Systems, San Diego, California, USA.