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

[Megaputers]

Dear Jakob,

Different data mining algorithms allow you to analyze data from 
different
angles. The best strategy is to use several algorithms, which would 
support
and validate the results of each other. You gain more insight in the
investigated issue when you use many independent analysis techniques.

I would say that you look for a specific answer to a very general 
question. I
tend to think that even a most vendor-neutral and 
insurance-experienced
analyst will have a really hard time trying to answer this question. 
The
answer depends very much on the objectives of your analysis, on the 
set of
variables for analysis available at ALKA, on the sheer amount of 
data, and so
on. Thus the solution to your task also will be very problem-specific.

I would like to suggest that you use a tool that provides not just 
one, but
several data analysis techniques if you wish to process your data 
in-house. In
addition, I would like to offer a cooperation of our experienced data 
analysis
team in working on your data. Megaputer has a good experience in 
marketing
applications. Recently we have been advising Prof. Raymond Burke 
through his
MBA course "Marketing Intelligence" at IU Kelley Business School. Some
specific examples can be found at
http://www.megaputer.ru/lessonse.html

Though we had analyzed some tasks there that sounded similar to your 
case, we
have learned that no two tasks have the same solution, no matter how 
similar
they might appear. Each problem has its own set of winning tools - a
combination of them most often. I will be happy to help finding the 
best one
for your case.

Sincerely,

Sergei Ananyan
Megaputer Intelligence, USA
812-325-3026 tel
812-339-1646 FAX
http://www.megaputer.ru

======================================================
<< I am trying to find out whether there in practice have been made 
any
comparison of different data mining methods with the purpose to 
predict
damages in an insurance company respectively to predict the 
additional sale
and drop-out behavior as to direct marketing activities.
 
 I am thinking of a comparision of e.g.
 
 · Logistic regression
 · Linear regression
 · Non-linear regression
 · Neutral net
 
 Which methods give notoriously the best stability as to a temporal 
point of
view.
 
 What is the difference in the different methods ability to predict 
the
customer behavior.
 
 Which method is the easiest to survey the effect of?
 
 If you have any practical experiences in comparisons in either 
insurance or
in marketing activities, I would be pleased to hear from you.
 
 
 Yours faithfully
 
 
 Jakob Laursen
 
 
 Head of dep. f Analysis and Support
 
 ALKA Insurance
 Engelholm alle 1
 Dk 2630 Taastrup
 Denmark >>