Decision/Regression Tree is the Most "Balance" Tree when Reduction in Entropy/Variance has the Highest Value

Course Q&A Machine Learning Specialization Advanced Learning Algorithms
1

Hi there,

When picking a split for a decision tree / a regression tree, I notice one thing and am not sure if it just so happened that it would appear the most balance when its reduction in entropy/variance has the highest value. If this was not coincidental, I wondered if this was “by design” and could be proven mathematically.

Appreciate your input.

Stanley

2

Hello Stanley, the only metric to consider for a split is maximum information gain, and any other interesting observations are just by-products.

Raymond

3

Thanks for your swift response, Raymond. Well noted.