Page 1 of 2
Student Name:________________________ Enrollment No.____________________
Comprehensive Examination
Course Name: Predictive Modeling for Data Science Course Code: DS 432
Academic Year: 2024-25 Semester: Jan-June (Even)
Date of Examination: 14-May-2024 Total pages: 02
Time: 10.30 AM to 01.30 PM Total Marks: 40
Instructions: 1. Attempt all the questions
2. Use of calculators in allowed
3. Assume data and draw digrams, if required.
Q. Questions Total COs
No. Marks
1 Explain the fundamental difference between supervised and unsupervised learning in the 2 CO1,CO2
context of predictive modeling. Provide one real-world example for each.
2 A regression model is evaluated on a test set with the following actual and predicted values 4 CO1,CO2
for the target variable:
Instance Actual Value Predicted Value
1 25 23
2 30 32
3 40 38
4 20 22
5 35 34
Calculate the following performance metrics:
a) Mean Absolute Error (MAE)
b) Mean Squared Error (MSE)
c) Root Mean Squared Error (RMSE)
d) R-squared (R2)
3 Define the null hypothesis (H0) and the alternative hypothesis (Ha) in the context of 2 CO1,CO2
hypothesis testing. What is the primary purpose of using ANOVA?
4 Given the Decision Tree, answer following questions: 4 CO1,CO3
a. Suppose a 25-year-old with an income of $60K is evaluated by this tree. Which path does
the data point follow, and what is the predicted outcome?
b. If a 35-year-old with any income is evaluated, what decision does the tree make?
c. How does the tree visually demonstrate the importance of "Age" compared to "Income"?
d. If you wanted to explain to a stakeholder why a 28-year-old with a $40K income is
predicted as "No", how would you use the tree diagram?
5 Explain the concept of cross-validation and why it is important in evaluating the 4 CO3,CO4
generalization performance of a predictive model. Describe common cross-validation
techniques.
Page 2 of 2
6 A marketing team is building a model to predict whether a customer will click on an online 4 CO1,CO2
advertisement (1 for click, 0 for no click) based on their age. They have trained a logistic
regression model and obtained the following equation for the probability of a click:
a) For a customer who is 30 years old, what is the probability they will click on the
advertisement according to this model?
b) Interpret the sign of the coefficient for Age (-0.02) in the exponent of the logistic
function. What does it suggest about the relationship between age and the
probability of clicking on the ad?
c) If the marketing team wants to target customers who have a probability of clicking
greater than 0.6, should they target younger or older customers based on this model?
Justify your answer.
7 You are given the following model to predict monthly Electricity Bill (in $) for a 4 CO1,CO2
household:
Bill=450+0.8(Units)+16.72(AC Usage)+0.78(Units*AC Usage)
Where:
• Units = number of electricity units consumed
• AC Usage = 1 if AC used frequently, 0 otherwise
a) Calculate Bill for a household using 1500 units without frequent AC use,
b) Calculate Bill for a household using 800 units with frequent AC use
c) Explain the interpretation of the interaction term coefficient (0.78).
d) Calculate the marginal effect of using an additional unit of electricity when AC is used
frequently.
8 You performed PCA on a dataset and obtain the following explained variance ratios: 4 CO1,CO2
Principal Component Variance Explained (%)
PC1 65.36%
PC2 26.36%
PC3 6.28%
PC4 2%
a) How many principal components are needed to retain at least 95% of the variance? (1
mark)
b) If your original dataset had 12 features, how many features are eliminated when using the
selected components? (1.5 marks)
c) List one drawback of using PCA in predictive modeling. (1.5 marks)
9 Explain the fundamental architecture of a basic artificial neural network (ANN), including 4 CO1,CO3
the roles of the input layer, hidden layer(s), and output layer. Describe the process of
forward and back propagation in detail, including how activation functions are used.
10 What is DBSCAN, and how does it differ from K-means and hierarchical clustering? 4 CO1,CO2
Explain the key parameters. Describe the three types of points in DBSCAN: core, border,
and noise points. What are the advantages and disadvantages of using DBSCAN for
clustering?
11 Explain Transfer Learning in the context of CNNs. 4 CO1,CO3
• What is the concept of transfer learning?
• How is it implemented with pre-trained CNNs (e.g., VGG16, ResNet)?
• What layers are typically frozen or fine-tuned?
• Advantages in terms of training time and data requirements
*****