One Step to Make Decision Trees Produce Better Results

Background, implementation, and model improvement

Decision trees (DT) get ditched much too soon.

It happens like this:

The DT is trained. Natural overfitting presents. Hyper-parameters get tuned (unsatisfactorily). Finally, the tree is replaced with Random Forest.

While that may be a quick win for performance, the replacement prioritizes a “black box” algorithm. That’s not ideal. Only a DT can produce intuitive results, offer business leaders the ability to compare trade-offs, and gives them a …

author business strategy data science decision decision-tree decision trees exploratory-data-analysis implementation natural overfitting parameters performance photo python random replacement tree trees