How to determine the most important factor for a particular scenario?

decision_trees

#1

the question was confuse me some days .

one day, the orders decline 1 million base on a week ago, and we have lots dimensions to analyse the decline, such as area,product,fee,sales type and so…

so how to determine which dimension was the most important dimension that have driven the decline?

and i have try use decision tree, however , didn’t performance good.


#2

Hi @will_xi,

Could you be more clear with the question? What is the problem statement?

If I am not wrong, you are trying to look for feature importance. Fit a random forest model, it provides an inbuilt command to determine the feature importance. Feature importance will tell you which feature from the complete dataset was the most important in making the predictions.


#3

sure,in daily, usually we need analyst what happened driven the decline, so often there must happen an unknown incident .

normally ,we analyst all the dimensions , these dimensions were related, such as area\product, they maybe have same trend. so how to determine which dimensions was the real factor driven the decline and have cause other dimension decline in the same time ?

i have use the decision tree with std. rep the gain value, however i feel the result was base on the structure of the dimension, just as one dimension have eight keys, then the std. must bigger than the dimension that have three keys.