I was watching Lesson 7 about the implementation of RandomForest, then at the moment we talked about finding the best split, Jeremey said we can use any scoring function, like RMSE or STDDEV each element in the two groups.
I was wondering if we can use a function that could calculate the homogeneity of a group (i.e, f(row1, ...., rown)) and choose the best split as the one generating groups with best homogeneity score.
This way:
- the leaves becomes a cluster of points and
- the DecisionTree will provide a hierarchical view