11-18-2014 04:57 AM
i am suspicious that some of my inputs are nearly same with my target variable. Because in desicion tree process these inputs receives considerably high logworth stats.
do you know how can i check correlations between my target and inputs?
11-18-2014 11:36 AM
I think what you need is to calculate VIF to assess multicollinearity. HPReg node this automatically for you. If you are working with an older EM version you can also write some code for proc reg. See https://communities.sas.com/message/188233
Since you are talking about trees, they are useful to detect 2-way interactions. After your first split, what variable is selected for the largest tree of maximum depth 2?
Do you get similar results from both VIF and a tree to detect 2-way interactions?
I hope that helps,
11-19-2014 04:31 AM
unfortunately i am working with a sas miner screen without a hpreG node. On the other i dont know how to write code in saas miner. By the way i have read the message you metioned
11-19-2014 12:24 PM
You can use the StatExplore node (on the Explore tab) to calculate correlations between an interval target and interval inputs.
11-20-2014 04:27 PM
Yes in that case you can use the Chi-Square Statistic reported in StatExplore. Usually it is used in the opposite way though - you want to keep the inputs associated with the target