Скачать с ютуб [MXML-2-09] Decision Trees [9/11] - CART, Cost Complexity Pruning (CCP) в хорошем качестве

[MXML-2-09] Decision Trees [9/11] - CART, Cost Complexity Pruning (CCP)

This video was produced in Korean and translated into English. And the voice is generated by AI TTS. The English translation may contain grammatical errors. In the last video, we looked at the feature importance and determined the optimal depth of the tree through cross-validation. And we tried using scikit-learn's DecisionTreeClassifier. In this video, we will look at the post-pruning method using Cost Complexity function. As the depth of the tree increases, errors measured by the training data may decrease, but errors measured by the test data may increase. This is a typical overfitting phenomenon. To prevent overfitting, you can prune branches from a complex tree to create a less complex tree. This is called the pruning technique. The pruning method using Cost Complexity corresponds to post-pruning. In the previous video, we set the max_length of the tree in advance and tried to find the optimal depth through cross-validation while changing the depth of the tree. This can be said to be a pre-pruning method. Post-pruning uses a method of creating the tree deep enough to prevent errors initially, and then removing unnecessary branches one by one later. The pruning algorithm by Cost Complexity was proposed by Leo Breiman and others when they proposed the CART algorithm in 1984. In this video, we will look at what the Cost Complexity Pruning is, and in the next video, we will implement the CPP using scikit-learn's DecisionTreeClassifier. #DecisionTree #CART #Pruning #CostComplexityPruning #PostPruning #CCP