I am trying to build a model that predicts a measure
(insurance claims, etc) against dimensions and member
properties (city, state, coveragetype, etc) and it seems
the mining algorithms can't function in this way because
the claim instances are all basically unique values,
which means it can't classify and segment them and show
patterns. I have manually segmented the values into
ranges, with limited success. I seems the models can
only predict the why (member properties, dimensions) and
not the what (measures), unless you classify the measures
by ranging them off. Has anyone else found a way to make
it work this way?

Hi Jim,

I guess that you're trying to predict the dollar amount of a claim based on
some other information (like demographics, specific coverage details, etc.).
If you define the predictable column as discretized, the algorithm will
create the buckets (segements) for you. If you want to be more granular
(narow ranges) you can specify for more buckets.

Hope this helps...

--
Raymond Balint [MS]
This posting is provided "AS IS" with no warranties, and confers no rights.



"Jim" <jim_callison (AT) hotmail (DOT) com> wrote