Dear Cristiano Mauro Assis Gomes

Essentially, cost complexity pruning adds another way to regulate the growth of a tree. This pruning strategy is specified in DecisionTreeClassifier by the cost complexity parameter, ccp alpha. Greater ccp alpha values increase the number of pruned nodes.

In Contrast, Pruning minimizes the final classifier's complexity, which increases predicted accuracy by reducing overfitting. While it decreases the size of a Decision Tree, which may somewhat raise your training error but greatly reduces your testing error, making it more adaptive.

Kind Regards

Qamar Ul Islam

Dear Qamar,

thank you for your answer.

I made available in my page in researchgate some published papers written by me about regression and decision trees, the cost complexity criterion, as well as the rationale of pruning trees. Most of these papers explain, in a didactical approach, what is cost complexity criterion and the rationale of pruning. If you have time and interest, you can read them. Hope you enjoy them.

Concerning my question, we have tested the efficacy of cost complexity criterion when pruning certain trees. We tested it through simulations; our results showed that this criterion was not effective for our data and our trees.

Our initial evidence show the importance of testing this criterion when applying it in trees.

If you have time and interest, you can see our analyzes and results through the link which I made available in my previous message. Hope you enjoy.

Best regards,

Cristiano.