Use of modern heuristics to transform and select regressors for linear modelling

Hi,
I am working on a project that intends to investigate the
implementation of a modern heuristic (e.g. simulated annealing,
genetic algorithms or local search) to search through a space of
polynomial transformations and assign selections for a linear
regression.

I have read that standard statistical methods for finding suitable
transformations of regressors use hill-climbing algorithms to search
for the correct transformations for linear modelling. I have found
that alot of times techniques such as stepwise regression have been
used to select a subset of regressors using a greedy algorithm.

BUT when this technique is used on a more complex model these
algorithms would fail to reach a global optimum.

I would like to know if by adopting a heuristic technique it may be
possible to provide better results.

(Could anyone post any suggestions/possible reading material/anything
that has been done along the same lines)

Thankyou,

[ comp.ai is moderated ... your article may take a while to appear. ]
.

Relevant Pages