SAP PA

Chapter 14: exercise 1

From data to decisions

Metadata:

default

(0 = no, 1 = yes) Has the customer defaulted on his/her loan?

checking account

Status of existing checking account

duration in months

Duration of the loan

credit history

How have the customer paid loans in the past?

purpose

Purpose of the loan

credit amount

How much is the loan?

savings account

Does the customer have a savings account and for how much?

employment

Employment status

installment rate

Installment (payment) rate as a percentage of disposable income

personal status and gender

Marital status and gender

other debtors

Are there other debtors and guarantors (co-signers)?

present resident since

Length of residence at current address (years)

property

What property does the customer own?

age in years

Age

other installment plans

What other loans does the customer have?

housing

Housing

# of existing credits at this bank

Other loans at this bank?

job

What job skills and status does the customer have?

# of people responsible for

How many dependents rely on this customer?

telephone

Does the customer own a phone?

foreign

Is this customer a foreign worker?

Build a classification model that classifies customers based on their default risk. Then answer the following questions.

1.What is the most important factor in predicting whether a customer will default? Does this make sense; is it logical that this is the most important factor? Why or why not?

2.What is the least important factor in predicting whether a customer will default? Does this make sense; is it logical that this is the least important factor? Why or why not?

3.What are the most important influencers for default prediction? Explain.

4.What data did you use for training and what data did you use for validation?

5.Show how well your model works on the validation data. Describe the model confidence.

6.Describe the confusion matrix. Explain the risk/loss for the bank when there are false positives and false negatives.

7.Can this model confidence be improved? If yes, how? What is the drawback of overfitting the training data?

8.Show the decision tree that classifies customers.

9.Show one recommendation for a specific customer and the variable values you chose for that customer.