none
feature importance for each row RRS feed

  • Question

  • I have used boosted decision trees for my model on flight delay prediction and deployed it as a web service. Now I want to find out which feature has been the most important/effective in prediction of delay for each input row for the test data. is it possible to do so in azure ml? permutation feature importance gives the score for the model only. is there any way to get score for features corresponding to each input row?

    • Changed type ayushi13 Wednesday, April 4, 2018 6:09 AM
    • Edited by ayushi13 Thursday, April 5, 2018 5:41 AM
    Wednesday, April 4, 2018 6:06 AM

All replies

  • There is! Try using the Permutation Feature Importance module. Once it's run, visualize it and it will give you a column of how important each column is. The values around 0 are what the algorithm deems as less important so you'll want to keep the columns that are higher.
    Wednesday, April 4, 2018 11:17 AM
  • I have already tried using permutation feature importance, but it gives the score for the entire model.

    what I want is to find the most important feature for each input row 


    ayushi13

    Thursday, April 5, 2018 5:41 AM
  • Hey, have you found the answer of this? Please share it here 
    Friday, September 20, 2019 5:14 AM