Earth: Multivariate Adaptive Regression Splines (MARS)

example2-earth-graph-grey.png example1-earth-graph-grey.png The earth package is an implementation of Jerome Friedman's Multivariate Adaptive Regression Splines, commonly known as "MARS".

The earth source code is licensed under the GPL and runs in an R environment, or can be used as a stand-alone C library.

Earth is derived from the mda:mars library written by Trevor Hastie and Rob Tibshirani.


Download link to CRAN

Standalone version for those who want to call earth from their own software without requiring R. A MATLAB interface is also provided (thanks to Fred Nicolls). This standalone version is based on earth version 3.2-6 (Nov 24, 2013).

General information on MARS

Wikipedia article (recommended)
Statsoft commercial vendor
Salford commercial vendor

Internal Details

Timing tests
Internal data structures
Internal call graphs

Citing earth

What people are using earth for

Earth has become quite popular. Here is a small sample of how people are using earth.

Max Kuhn at Pfizer is using earth for building predictive models in computational biology and chemistry. You may also want to check out Max's caret package.

Thorsten Behrens at the University of Tuebingen (Institute of Geography) is using earth for Pedometrics / Digital Soil Mapping.

Trevor Nell at Brolink (Pty) Ltd is using earth to analyse insurance claim data to determine appropriate premiums.

Kurt Salmela and Ming Shan of Maritz Research are using earth for analyzing marketing research survey data.

Darin Brooks at the University of British Columbia and Kim Forest Management Ltd is using earth for predicting secondary growth/mid-term timber supply in the BC forestry sector.

To Stephen Milborrow Homepage