Re: About wrong winner predicted in the model

Tech-Archive recommends: Fix windows errors by optimizing your registry

It seems to be a problem with presenting data to the model, not with the
model itself.

Let's assume (hypothetically, because I do not know the actual data) that
India played 20 games with Japan and 5 games with all other countries. Let's
also assume that Japan won 15 of those 20 games with Japan. The model
notices that Team1 is India and predicts, most likely, Japan is the winner
(because there are no other games between India and England, and based on
Team1=India, most victories were Japan's)

One possibility is to add a new column to your data, which has only 2
possible values: 1 (meaning Team1 won) or 2 (meaning Team2 won) . This
column should be used as target (predictable)
Another potential column would be City (1 if it is home city for team 1, 2
if it is home city for team 2), to be used as input

This way, the data that the model sees describes the differences between
team 1 and team 2 and the training is optimized for discerning between team1
and team2.

Could share the list of columns (or, even better, a small piece of the data)
?


Hope this helps!
--
--
--
This posting is provided "AS IS" with no warranties, and confers no rights.
Please do not send email directly to this alias. It is for newsgroup
purposes only.

thanks,
bogdan

"Tong" <Tong @discussions.microsoft.com> wrote in message
news:7FB1F15A-A4F4-4028-BFF0-B3E726DEBA13@xxxxxxxxxxxxxxxx
I am using data mining of SQL server 2005 for one sport as basketball. I
have
thousands of match records of this sport. All the data points are in one
table. In each record the most important columns are team1 and team2 for
the
two teams, and winner that tell us who win this match. There are several
other columns that are about these two teams. I set up models using Neural
Network and Decision Trees. After process the models, I use them to tell
me
who the winner will be. Though most results are one of team1 and team2,
there
do be a few results that are not team1 or team2. For example, for one
match
between England and India. The model will predict winner will be Japan. Is
it
an error of my work? Or this is normal for this kind of prediction?
Another problem is how I can tell the computer some columns are for team1,
and some other columns are for team2, or just let the computer do the
model
process.
Thanks a lot for any advice.

Tong



.

Relevant Pages

  • Re: About wrong winner predicted in the model
    ... India played 20 games with Japan and 5 games with all other countries. ... notices that Team1 is India and predicts, most likely, Japan is the winner ... In each record the most important columns are team1 and team2 for ... and winner that tell us who win this match. ...
    (microsoft.public.sqlserver.datamining)
  • Re: Bidirectional Binary Self-Joins
    ... constraints. ... Unfortunately we are left with that same issue - the choice of team1 ... and team2 is arbitrary, and as David or Paul pointed out, it means I ... I would actually add a surrogate identifier for the game if I had to ...
    (comp.databases.theory)
  • About wrong winner predicted in the model
    ... Another problem is how I can tell the computer some columns are for team1, and some other columns are for team2, or just let the computer do the model process. ... Tong ...
    (microsoft.public.sqlserver.datamining)