RE-EM Trees: A Data Mining Approach for Longitudinal and Clustered Data

This page contains material related to the paper "RE-EM Trees: A Data Mining Approach for Longitudinal and Clustered Data," by Rebecca J. Sela and Jeffrey S. Simonoff, which appeared in Machine Learning in 2012. The paper discusses the implementation of CART-based recursive partitioning (trees) to longitudinal data through a mixed effects model framework.

Click here to access the paper on the journal's web site.

Click here for supplemental diagnostic plots related to the linear mixed effects and REEM tree model fits to the software transactions data.

An R package to fit the RE-EM tree is available at CRAN. The current version of the package is 0.90.4, and any earlier versions should be updated or replaced. The package produces an object that includes a Tree attribute that contains the fitted tree; this can be plotted using the packages rpart.plot (via the command rpart.plot(mytree$Tree, extra=0, fallen.leaves=FALSE)) or partykit (via the command plot(as.party(mytree$Tree), type="simple")).

Note that a function is also available to implement conditional inference-based trees to longitudinal data; such trees have the property of being unbiased in their selection of splitting variables. See this page for more details.