Classification and Regression Tree

Question 1

How to create classification and regression tree?

Answer 1

• )Build a tree by splitting on variables
• )Follow the split and predict the most frequent outcome
• )Interpretable
• )Does not assume a linear model

Question 2

When does CART stop splitting?

Answer 2

One way is to setting lower bound. R uses minbucket parameter

Question 3

General formula for creating tree

Answer 3

library(rpart)
model = (Y~X, data = train, method = “class”, minbucket = 25)
predictions = predict(model, newdata = test, type = “class”)

table(test$Y, predictions)

Question 4

Explain sensitivity

Answer 4

TP / (TP + FN)

What percentage of True cases correctly identified from all True cases.

Question 5

Explain Specificity

Answer 5

TN/(TN+FP)

What percentage of False cases correctly identified form all False cases.

Question 6

What is bagged/bootstrapped sample of data?

Answer 6

each sample used to create a model select randomly with replacement.

Question 7

What is K-fold cross validation?

Answer 7

split training set into k pieces. use k-1 pieces to estimate the model and kth model (validatioan model) to test each parameter value.

Question 8

What is CP value?

Answer 8

cp is complexity parameter. measure trade off between model complexity and accuracy on training set.