Multivariate feature selection with hybrid heuristic search approach.
Feature selection is driven by Hotelling-Lawley Trace statistic (parametric).

Main usage is likely the hybridFeatureSelection function, which takes:
- Data frame of predictors as x
- Vector / column of class(es) as y
- 'start' parameter is method to select first variable for the search (can be 'random' or 'Hotelling')
- stopP parameter is number of features to select before ending search
- stopT2 parameter specifies Hotelling-Lawley trace statistic to reach before ending search
(meeting either stopP or stopT2 will end search)
The function performs a heuristic search to find the best (local) optima until it hits one of the stop parameters. 

There is also a modified bagging routine to obtain out of bag estimates for features including LDA classification accuracy for a 2 class scenario (perhaps additional class scenarios will be supported at some time).

Code by Jyota Snyder (jyotas at my dot ccsu dot edu).
Concepts, approach, hybrid feature selection, and modified bagging schema were adapted into R from Darius Dziuda's 2010 "Data Mining for Genomics & Proteomics" book.