To improve the prediction performance we propose new preprocessing method
for spreadsheet formatted data.  This method decomposes input data into
reusable parts, each of which contains a rule.   We firstly discuss the
Matchable principle, which supporting generalization ability.  This
principal emphasizes the increasing of matching chance to extract
information structure.  Next we derives a Matchable criterion suitable for
the situation decomposition from the principal.

Then we developed a search algorithm for decomposition and speed up it by
reducing enormous search space.   Simulation demonstrates that the
algorithm can decompose mixed situations. This technology is effective for
pre-processing of data analysis and pattern recognition.

Keyword: Information structure, Matchability criterion, Situation
decomposition, Rule extraction, Prediction performance, Data mining