|
I know the process of growing a decision tree when the data points are not weighted using information gain method. So suppose I have 1000 data points in the training set, each represented by 10 features and in addition let's say I have weights assigned to eah of those data points, what is the method to grow the tree? In what step of the decision tree should I consider the weights of the individual data points and how can I do that? |
|
Your question has been partially answered on this topic: http://www.csresearchers.com/?qa=12/using-sample-weighting-with-svm-decision-tree-classifiers It seems that you should use a weighted version of the mutual information or entropy that you are using for the split in the decision tree classifier. Thanks, now it is clear for categorical attributes that we may want to weigh the mutual information or entropy. In case of regression where we have to decide the split on an attribute that is continuous, how do we consider the weights?
(Oct 31 '14 at 05:46)
Abhishek Shivkumar
|
