Wanted to make sure that I have the proper steps for building a Regression Decision Tree correct. Please let me know if I'm missing anything or if something looks incorrect:
- Build a fully grown tree:
- Divide the predictor space (all X’s) into distinct and non overlapping regions. Where the regions is split is determined by the maximum decrease in RSS. RSS measured by (actual – predicted), and predicted is always the Mean of the train datapoints in whatever region.
- Repeat until a stopping criteria. Eg. N<5 in each region.
- Set up the tree to be pruned to prevent overfitting.
- Vary Alpha (basically Lasso’s Lambda but for Trees) to create subtrees with different numbers of terminal leaves.
- Do K fold CV to pick the right Alpha. Tree size with the Lowest MSE wins.