Search based software enigneering is a new problem formulation
in the field of software enigneering that is intended to
bring a new problem solving technique to the field,
metaheuristic search.
This thesis is an empirical study of eight different
models that demonstrate the capibilities of
Iterative Treatment Learning (ITL),
a machine learning based metaheuristic search technique.
ITL differs from other mainstream search techniques
by not offering value assignments to all 
search dimensions.
This gives the solution found by ITL inherently more
flexibility, which can be of great value
when used in early project planning.
ITL has been previously introduced by Menzies et.\ al.
These previous studies used a smaller number of less
complex models and did not explore a range of options to ITL.
Menzies \etal also failed to realize the similarities between
ITL and metaheuristic search.
The metaheuristic search metaphor is particularly
insightful when describing possible future work.

The current work also introduces a new discretizer tailored 
to work with ITL, called extreme sampling.
Extreme sampling readily outperforms the previous discretizer,
diagonal striping.
This new discretizer allows ITL to find solutions
of the same quality as another highly regarded 
search technique, simulated annealing,
but an order of magnitude faster.

Using ITL in a new model development and execution environment, SPY,
allows us to find property violations in temporal models
with real valued inputs.
SPY is also compared to a commercially available tool, Reactis,
that uses random and heuristic search to attempt
to find property violations in temporal models
with real valued inputs.
SPY's learning capabilities are explored with 
biomathematical models written by hand and 
NASA flight models translated from another modeling language.