I ~ N. L..
Dalmia®
~ ~ ~ Institute of Management
\, ~ Studies and Research
~ (A School ,!f&cellence ofN. L. Dalmia EJ11cational Society)
A. Y. 2025 - 26
Course: PGDM-BA I
Batch : 2024 - 26 Trimester V
Div. G Marks so
Date: 23 rd December 2025 Time: 2 hrs 30 mins
Subject: Machine Learning Techniques Subject Code : PGDBAS13
INSTRUCTIONS:
I. Question No I is compulsory and carries 20 marks.
2. From the remaining 4 questions answer any 3. Each question carries 10 marks.
Q. No. 1 Case Study or any other situation/scenario based question 20 Marks (C03,4)
A fast-growing digital lending platfonn wants to automate and optimize its loan approval process.
The goal is to:
1. Decide whether a loan application should be approved or rejected (binary
classification).
2. Assign a risk score to each applicant between O and I to estimate the likelihood of
loan default. •
Available Data:
• Applicant demographics: Age, employment type, education level, income, and
dependents.
• Financial behaviour: Credit history, existing loans, overdue payments, and spending
patterns.
• Application data: Loan amount requested, loan purpose, repayment tenure.
• External data: Credit bureau score, regional economic indicators, social scoring
(alternative credit scoring signals) .
•
Challenges:
• Regulatory constraints: System must comply with fairness, explainability, and audit
requirements.
• Data sparsity: Some applicants (e.g., students, micro-business owners) have limited
historical financial data.
• Dynamic socio-economic trends: Economic shifts (inflation, recession) can change risk
patterns.
• Fraud and gaming risk: Some applicants may manipulate data or exploit weaknesses in
scoring logic.
Questions:
(a) Model Selection:
a) Which machine learning models would you choose to:
Approve or reject a loan application?
Page 1 of 3
~ N. L. Dalmia®
~ ~ ~ Institute of Management
~ 117 ~ Studies and Research
(A School of Excelltnct ofN. L Dalmia Educalional Society)
Predict a continuous credit risk score?
b) Would you use one combined model or separate models? Explain your reasoning
with respect to interpretability and perfonnance.
(b) Data Challenges:
a) How would you address missing or limited financial history for first-time borrowers?
b) Suggest feature engineering techniques to extract behaviour patterns from time-series
repayment or spending data.
(c) Real-Time Prediction:
a) What approaches would you use to ensure fast predictions during high application
traffic?
b) How would you handle explainability requirements while maintaining computational
efficiency?
(d) Model Monitoring and Retraining:
a) How would you detect drift caused by economic fluctuatio~s or changes in user
behaviour?
b) Describe a retraining or reinforce1:I1ent strategy to ensure the model remains fair,
accurate, and compliant with regulations.
Q. No. 2. 10 Marks (COl)
A machine learning model has high accuracy on the training set but performs poorly on the test
set.
a) Identify the problem (overfitting, underfitting, or data bias) and explain its possible causes. (5
Marks)
b) Suggest methods to improve the model's generalization performance. Include regularization,
cross-validation, and data augmentation in your explanation. (5 Marks)
Q. No. 3. 10 Marks (CO2)
a) A health insurance company uses a logistic regression model to predict whether a customer
will file a medical claim within the next year ("yes" or "no"). The prediction is based on two
variables: ( 8 Marks)
1. Age (in years)
2. Annual doctor visits (count)
The logistic regression formula is:
z = BO + Bl .age + B2. visits
Where:
• BO= -3.8 (intercept)
Page 2 of 3
~ N. L. Dalmia®
-;; ~ ~ Institute of Management
'\;~~ Studies and Research
(A School of Excellence ofN. L. Dalmia Educational Society)
• Bl= 0.04 (coefficient for age)
• B2 = 0.25 (coefficient for doctor visits)
Calculate the probabilities of "yes" (the customer will file a medical claim) and "no" (the customer
will not file a claim) for a customer with:
• Age= 45 years
• Annual doctor visits = 5
b) Explain what stratify does in while making the train test split. ( 2 Marks )
Q. No. 4. 10 Marks (C03)
Explain the hieraJchical clustering technique and compare it with K-Means clustering with examples
Q. No. 5. 10 Marks (C04)
Using Random Forest as a reference classification machine learning algorithm explain
• Bagging and Boosting
Page 3 of 3