Re: DM: Comparision of Different Data Mining Methods in Insurance

[Ronny Kohavi]

Jakob> I am trying to find out whether there in practice have been
Jakob> made any comparison of different data mining methods with the
Jakob> purpose to predict damages in an insurance company
Jakob> respectively to predict the additional sale and drop-out
Jakob> behavior as to direct marketing activities.

Which car would you recommend to a 6"1' white male age 30 working
at a software company?

You may rule out a few cars, but at the end of the day you'd probably
want a *lot* more information about preferences and also suggest that
he test drive several cars.  The same holds for your question.

You're unlikely to find papers that will generalize to your
specific dataset, especially with respect to your question:
  What is the difference in the different methods ability to predict 
the
  customer behavior.
Different encoding of the same domain yield completely different
results on different algorithms.

These days it's not hard to load your data into a database
and run different algorithms to compare their performance.

Check out:
  Kohavi, R., Sommerfield D., Dougherty J., Data Mining using MLC++, a
  Machine Learning Library in C++. International Journal of Artificial
  Intelligence Tools, Vol. 6, No. 4, 1997, p. 537-566
available off:
  http://robotics.stanford.edu/users/ronnyk/ronnyk-bib.html
for this "test-drive" motivation.

--

   Ronny Kohavi (ronnyk@sgi.com, http://mineset.sgi.com)
   Engineering Manager, MineSet.
   Maximize the value of data with data mining and visualization.