Problem 1:

1.
Growing method: CRT
Validation: Holdout partitioning with 70% training and 30% testing split
Maximum tree depth: 10
Minimum cases for parent node: 20
Minimum cases for child nodes: 10
Impurity measure: Gini Index
2.
The final tree has 7 number of nodes and 4 terminal nodes.
3.

From the above chart we can see that V10, V11 and V1 are the top three major attributes in
the dataset to make a decision on splitting the tree.
4.
After increasing the minimum number of cases for the parent node from 20 to 25 and child
node from 10 to 15, the complexity of the tree decreases (from depth 3 to 2). The accuracy
of the training set increases from 91.1% to 91.8% and accuracy of test set decreases from
93% to 88.2%.
The decrease in the complexity can be explained as increasing the min number of cases in
child node, and as child not having enough cases hence, the parent cannot split further.

PROBLEM 2.
1.

With above data,
There are 6 classes in red wine data set. And we can say that dataset is slightly right skewed.