\bi
\item ``Tag'' each record with its cluster identifier
\item Using feature subset selection, learn which SILAP attributes
are most important
\bi
\item 10-way,  CFS,  selected:
{\color{red} HS2, EX3, US3, CL3, FR3, DT3, RM3}
\ei
\item Learn a decision procedure that identifies  each clusters
\item If SILAP performs differently for each cluster, then
those clusters represent truly different project types.
\ei
{\tiny
\begin{verbatim}
DT3= use of defect tracking; CL3= CMM level; US3= use of standards;
EX3= experience;             HS2= human safety

DT3 <= 1: cluster2 (150.0/4.0)
DT3 > 1
|   CL3 <= 4
|   |   US3 <= 2: cluster0 (170.0)
|   |   US3 > 2
|   |   |   DT3 <= 2
|   |   |   |   EX3 <= 2: cluster3 (25.0)
|   |   |   |   EX3 > 2
|   |   |   |   |   HS2 <= 3: cluster0 (14.0)
|   |   |   |   |   HS2 > 3: cluster3 (6.0)
|   |   |   DT3 > 2: cluster0 (42.0)
|   CL3 > 4: cluster1 (19.0)
\end{verbatim}
}
In a 10-way cross-val, accuracy=99.061\% (!!).

~\\

``The variables in the tree are all about how much a project knows
about itself and how much it is willing to share that knowledge
with others.''