Lets say in the training cases, 90 % of the product purchase was chicken and
only 3% was fruit. How can I accurately predict "fruit"? I have been tweaking
the parameters with differernt values and looking into the lift chart. Tried
to make sure
that the split is correct but still no success. Most of the predictions are
"chicken" even though in the newpurchase cases I have 4% fruit.
Please help and TIA...
[quoted text, click to view] "sqlster" wrote:
> I have been reading chapter on decision tree algorithm in sql server 2005
> data mining book.
> Even though it discusses the algorithm parameters very well, is there any
> other source where
> I could read up on algorithm parameters?
>
> Here is the case I am working on. I have a customer demographic table with
> about 8 attributes
> and I predict on what (grocery product : fruit, juice, chips, chicken) they
> would buy.
>
> Out of 400 cases, I take out 40 cases with known product choice and call it
> newpurchase.
> Another 100 cases, I separate out for testing as testpurchase and rest it
> for training called
> trainpurchase.
>
> I played with different attributes in decision tree model and ran it against
> newpurchase.
> 90% of predictions were correct and rest were wrong. I have been looking at
> case generated
> etc. etc. Is it possible that even though attributes point to the correct
> product in
> terms of statistics but customer chose a different product any ways????
>
> I could use a real good discussion on decision tree algorithm parameters and
> how it influences the growth/pruning of the tree.
>
> Please help and let me know if I need to further clarify this.
>
sqlster,
Check out books online for an excellent description on algorithm parameters.
http://msdn2.microsoft.com/en-us/library/ms175312.aspx HTH..
--
http://zulfiqar.typepad.com BSEE, MCP
[quoted text, click to view] "sqlster" wrote:
> I have been reading chapter on decision tree algorithm in sql server 2005
> data mining book.
> Even though it discusses the algorithm parameters very well, is there any
> other source where
> I could read up on algorithm parameters?
>
> Here is the case I am working on. I have a customer demographic table with
> about 8 attributes
> and I predict on what (grocery product : fruit, juice, chips, chicken) they
> would buy.
>
> Out of 400 cases, I take out 40 cases with known product choice and call it
> newpurchase.
> Another 100 cases, I separate out for testing as testpurchase and rest it
> for training called
> trainpurchase.
>
> I played with different attributes in decision tree model and ran it against
> newpurchase.
> 90% of predictions were correct and rest were wrong. I have been looking at
> case generated
> etc. etc. Is it possible that even though attributes point to the correct
> product in
> terms of statistics but customer chose a different product any ways????
>
> I could use a real good discussion on decision tree algorithm parameters and
> how it influences the growth/pruning of the tree.
>
> Please help and let me know if I need to further clarify this.
>
Don't see what you're looking for? Try a search.
