decision trees

Question 1

arrange the data into predefined groups

Answer 1

classification

Question 2

What is the difference between clustering and classfication?

Answer 2

depends whether categories are predefined or not

Question 3

What are examples of classification

Answer 3

classifying emails as “legit” or “spam”

Question 4

what are common algorithms of classification?

Answer 4

decision tree analysis

Question 5

What is the processes of how classficiation works

Question 6

What is the processes of how classficiation works

Answer 6

1) choose classes and a set of classifying attributes
2) choose a set of records (randomly) for the training set
3) choose a set of records (randomly) for the test set
4) using the training set, create a model to predict the chosen class as a function of the other classifying attritbutes
5) evaluate the model using the test set
6) classify future records by applying the “best” model

Question 7

What attributes of families made it more likely they rented a certain class of apartment?

Answer 7

To answer this, we can run a classification algorithm over the training data
- the data has known outcomes, we know what class of aprtment each family ended up renting
- the outcome can be displayed on a decision tree

Question 8

The classifcation of events, outcomes, things etc

Answer 8

Decision tree

Question 9

what is the first question of the decision tree called

Answer 9

root node

Question 10

What is the second question of decision tree called

Answer 10

split or partition

Question 11

what is the third or ifnal part of decision tree called?

Answer 11

terminal node or lead

Question 12

How do you create a decision tree using training data?

Answer 12

find a way to split (partition) the training dataset into smaller sub-groups

Question 13

How to decide what attributes and rules to split at each node? (which partition is better than others?)?

Answer 13

Several algorithms. One appraoch: recursive paritioning

Question 14

What are the steps to recursive partitioning steps?

Answer 14

1. pick one of the predictor variable, Xi
2. Pick a value of Xi (says, si) that dvides the training data into two portions
3. measure how “pure” each of the resulting portions (subgroups) are
4. The idea is to pick Xi and Si to maximize purity improvement in one step
5. REPEAT the process for each fo the subgroups until a predetermined number of subgroups is reached, or until improvements in purity become too small

Question 15

What does pure mean?

Answer 15

containing records of mostly one class

Question 16

What occurs at each split?

Answer 16

the algorithm maximizes the purity of the resulting nodes
- a node is 100% pure when all of its data belongs to a single class
- A node is 100% impure (0% pure) when a node is split evenly 50/50

Question 17

what is the objective of producing a decision tree verizon?

Answer 17

split the training population to maximize overall purity gain

Question 18

I(N) is the IMpurity of

Answer 18

node
- formula is on slide 19 of decision trees powerpoint

Question 19

What is overfitting?

Answer 19

when the model fits the training data so well that it doesn’t do well with new data

Question 20

how to reduce the risk of overfitting

Answer 20

-pruning
-specifying a minimum bucket (node) size

Question 21

what is pruning

Answer 21

it reduces the size of decision trees by removing splits that provide little power to classify instances