Quiz 1

You might also like

Download as docx, pdf, or txt
Download as docx, pdf, or txt
You are on page 1of 2

Quiz 1

Full Marks 10

1. When a credit card user applies for increase in credit limit to a requested amount, the
bank uses different information like CIBIL credit score (‘crs’), income bracket (‘es’) and
other factors for making a decision about the credit limit increase. The bank will take one
of the following three decisions (des):
increase to the requested amount (“Y”).
reject the increment request (“N”)
increase to an amount which is less than the request amount (“M”).

The predictor set is gender (categorised as “female” and “male”), income (categorised
as “low”, “medium”, “high”), CIBIL credit score (continuous, between 350-850).

Consider the following R output after using multinomial logistic regression model:

Coefficients:
(Intercept) eslow esmiddle crs
N 2.90 -0.038 0.38 -0.006
Y -1.45 -1.03 -0.60 0.005

Suppose, ‘eslow’ represents the indicator variable for low income user, ‘esmiddle’
represents the indicator variable for middle income user. Define, interpret the value 0.005
in the above output. (2)

The log odds of increasing the credit limit to the requested amount vs. increasing the
credit limit to the requested amount for an individual will increase by 0.0005 if credit
score improves by 1 point.

2. Suppose you are using logistic regression to model loan approval status (either loan will
be approved or not) from Bay of Andhra bank using a single predictor - CIBIL score
(continuous). In the R output, you found that the regression coefficient estimates came
out to be 5.52 (estimate corresponding to the intercept) and 0.51 (estimate corresponding
to the CIBIL score). On checking the p-value, you found that the
(a) Write the logistic regression model you will use to predict the loan status? (1)

log(p/(1-p))=5.52+0.51CIBIL Score

(b)How will you interpret the regression estimate corresponding to the slope parameter in
this context? (2)
As the CIBIL score increases by 1 unit, the log of odds of loan approval increases by
0.51.
(c) How many different ML algorithms you know which can act as competitors to this
model? (1)
Probit Regression, Complementary log-log Regression, Classification Trees, Bagging,
Random Forest.

3. A dataset consisting of 1000 credit card holders was considered. This dataset was a result
of a market research survey which asked these credit card holders about the total number
of hotel bookings they made using credit card in the past three years, monthly credit card
bill, .

(a) Which ML model you should apply to predict the number of hotel bookings?
Explain your choice. (2)

The response variable is the number of hotel bookings, which is count data.
Hence, we can apply Poisson regression model.

(b) Can we use classification tree or a pruned classification tree or Bagging? Explain
in details. (2)
The response variable is counting in nature. So, one should not use a
classification tree or a pruned classification tree without converting the count
data. But the company should agree with the conversion of count data as the
conversion may change the meaning totally.

Bagging method can however be used as the method can be applicable to any
problem.

You might also like