EWStools is the source-code behind many of the modules within the Wisconsin
Dropout Early Warning System created
by the Wisconsin Department of Public Instruction. While
the framework was designed particularly to the development of early-warning
predictive models on education data, these tools represent a more generalized
framework for building, testing, and exploring models built through the train
function in the R package caret. As such, this package extends the features
of caret to make it more efficient to search across model types, explore
model performance on test and training data, and to draw ROC comparisons of
classification models specifically.
EWStools is currently in beta and many of the functions are changing regularly.
EWStools provides two distinct sets of features for model builders. The first
is tools to automate the search for the best fitting model across model
types. The second set of features is the creation of a new object class, ROCit
objects, which allow for the easy comparison of ROC performance of
classification models on both test and training data.
EWStools features wrapper code for caret's train function which makes it
easy to build a sequential test of many model types available to train and
store the results of the test efficiently.