K.S.

RANGASAMY COLLEGE OF TECHNOLOGY,TIRUCHENGODE - 637215

ECE
(An Autonomous Institution)

QUESTION BANK
Programme & Branch B.E & Electronics and Communication Engineering Semester VI
Course Code & Name 60 EC 604 & Machine Learning Techniques

Unit - I

PART - A Mar Blooms CO PO

S.No. ks level
1. List the types of activation function used in machine learning. 2 Re CO1 PO1
2. What does the bull’s-eye diagram represent in machine learning? How does it 2 Un CO1 PO1
relate overfitting and underfitting?
3. What is the mathematical representation of a perceptron model? 2 Re CO1 PO1
4. Why is ReLU preferred over sigmoid and tanh in machine learning for medical 2 Re CO1 PO1
imaging?
5. Compare validation set and test set in training process. 2 Un CO1 PO1
6. What is gradient descent rule in neural network? What is its use? 2 Re CO1 PO1
7. If a model exhibits high bias, what adjustments can be made to improve its 2 Ap CO1 PO1
performance?
8. Tabulate the difference between gradient descent and stochastic gradient 2 Un CO1 PO1
descent.
9. Define bias. Why is it needed in machine learning? 2 Un CO1 PO1
10. How generalization helps to improve the model in machine learning? 2 Un CO1 PO1
11. What is overfitting? How does it relate to the theory of generalization? 2 Ap CO1 PO1
12. Write the formula for sigmoid, hyperbolic tangent and ReLU activation function 2 Re CO1 PO1
and give its graphical representation.
PART- B Mar Blooms CO PO
S.No.
ks level
1. a. List the different machine learning types. Give a real life example for each Ap 8 CO1 PO1
category and state how machine learning is being used in any one types with an
example.
b. Discuss the bias-variance tradeoff and its significance in model selection. Un 8 CO1 PO1
2. a. Implement the two input AND & OR Boolean functions using perceptron model Ap 8 CO1 PO2
and discuss why a single perceptron model is not suitable for implementing XOR
functions.
b. What are bias and variance? Explain the concept of overfitting and underfitting Un 8 CO1 PO1
and state its effect on model performance.
3. a. Draw a single layer perceptron model with all its fundamental parameter and Un 6 CO1 PO1
explain how it helps to do linear classifier with an example.
b. Consider a neural network with one input layer, one output layer, and one hidden Ap 10 CO1 PO2
layer. The initial network has 10 neurons in the hidden layer. Describe how
increasing the number of neurons in the hidden layer might impact the model's
performance. Provide insights into the trade-off associated with choosing a larger
hidden layer size with necessary expressions.
4. a. Discuss back propagation algorithm with relevant equations and diagram, explain Un 10 CO1 PO1
 how it helps to reduce the error.
b. A 4-input neuron has weights 1, 2, 3 and 5. The transfer function is linear with the Ap 6 CO1 PO2
constant of proportionality being equal to 3. If the inputs are 2, 10, 5 and 20
respectively, what will be the output of a neuron?
5. a. Draw the basic architecture of a multi-layer perceptron and explain how it solves a Un 6 CO1 PO1
non-linearly separable problem.
b. Consider a neural network with one input neuron, one hidden layer with two Ap 10 CO1 PO2
neurons, and one output neuron. The network uses the sigmoid activation
function. Given a set of input features, calculate the forward pass, compute the
error using mean squared error (MSE), and then perform one iteration of the
backpropagation algorithm to update the weights.

Input features: [0.5]

Target output: 1
Learning rate: 0.1

Initial weights and biases:

W ih1 = 0.8
W ih2 = - 0.4
bh1 = 0.2
bh2 = - 0.1
W ho1 = 0.3
bo = 0.5
6. a. Discuss the role of activation functions in neural networks. Elaborate it. Un 6 CO1 PO1
b. Illustrate with a suitable example the cost function when age, height and gender of Un 10 CO1 PO1
a person would affect his/her weight. (MAY - 2022)
UNIT- II
PART- A Mar Blooms CO PO
S.No.
ks level
1. Differentiate Linear regression and Logistic regression. 2 Un CO2 PO1
2. What are support vectors? How it helps to do model performance? 2 Re CO2 PO1
3. What is large margin classifier? Why is SVM considered as large margin 2 Ap CO2 PO1
classifier?
4. Justify with a suitable example as when and where ‘Regression’ can be used in 2 Un CO2 PO1
a Machine Learning problem.
5. What is the significance of kernel functions in SVM? List any two kernel 2 Un CO2 PO1
functions.
6. The input image has been converted into a matrix of size 28 X 28 and a 2 Ap CO2 PO1
kernel/filter of size 7 X 7 with a stride of 1. What will be the size of the
convoluted matrix?
7. Define a kernel in machine learning. State its use. 2 Un CO2 PO1
8. Differentiate a linear SVM and a non-linear SVM. 2 Un CO2 PO1
9. What are the applications of support vector machine? 2 Un CO2 PO1
10. What is the primary purpose of a Classification and Regression Tree (CART)? 2 Un CO2 PO1
11. How is logistic regression used for binary classification? 2 Ap CO2 PO1
12. What is the significant use of Radial Basis Functions (RBF) in machine learning? 2 Re CO2 PO1
S.No. PART- B Mar Blooms CO PO
ks level
1 a With the help of regression line equation, explain how you will find a best fit line in 8 Un CO2 PO1
terms MSE and residuals.
b For the following set of training samples, find which attribute can be chosen as 8 Un CO2 PO2
the root for decision tree classification:
Instance Classification A1 A2
1 + T T
2 + T T
3 - T F
4 + F F
 5 - F T
6 - F T
(MAY – 2022)

2 a Compare Lasso regression and Ridge Regression in terms of bias and variance. 6 An CO2 PO1

b Suppose you have a dataset with two classes that are not linearly separable. 10 Un CO2 PO2
How would you use a support vector machine (SVM) to classify this data?
Discuss the use of kernel functions and the choice of parameters in this scenario.
3 a Study the relationship between the monthly e-commerce sales and the online 10 Ap CO2 PO2
advertising costs. Here the survey result for 7 online stores for the last year was
given. Your task is to find the equation of the straight line that best fits the data
and plot the regression line for the sample values. The following table represents
the survey results from the 7 online stores.

b What is the purpose of regularization in machine learning? How do regularization 6 Ap CO2 PO1
parameters, such as lambda in L1 and L2 regularization, influence the trade-off
between bias and variance in machine learning models? Explain.
4 a Explain the fundamental principles of logistic regression and analyze how it 8 Ap CO2 PO1
compares to other classification algorithms (e.g., decision trees, support vector
machines) regarding performance, interpretability, and scalability - examples
where logistic regression is preferred over other methods.
b Given the positively labeled data points (4,1)(4,-1)(7,1)(7,-1) and negatively 8 Ap CO2 PO3
labeled data points (2,0)(1,1)(1,-1)(0,0) in R2. Identify the support vectors and
discover a simple SVM that accurately discriminates the positive and negative
classes.
(MAY – 2020)
5 a Give decision trees to represent the following Boolean functions: 10 Ap CO2 PO2
(a) A ∧ ¬B
(b) A ∨ [B ∧ C]
(c) A XOR B
(d) [A ∧ B] ∨ [C ∧ D]
b Discuss advantages and limitations of Bayesian model compared to traditional 6 Un CO2 PO1
methods.
6 a A company manufactures an electronic device to be used in a very wide 8 Un CO2 PO2
temperature range. The company knows that increased temperature shortens the
lifetime of the device, and a study is therefore performed in which the life time is
determined as a function of temperature. Express the lifetime as a linear function of
temperature. The following data is found:

Temperature 10 20 30 40 50 60 70 80 90
(C)
Lifetime 420 365 285 220 176 117 68 34 5
(hours)
(MAY – 2022)
 Person Height Weight
1 167 55
2 120 32
3 113 33
4 175 76
5 108 25
b Discuss the core principles underlying Support Vector Machines (SVM) in 8 Un CO2 PO1
machine learning and how they enable classification in high-dimensional spaces.

Person Height Weight

1 167 55
2 120 32
3 113 33
4 175 76
5 108 25

UNIT - III

PART- A Mar Blooms CO PO

S.No. ks level
1. How does the K-means algorithm work? 2 Un CO3 PO1
2. What are the limitations of the K-means algorithm? 2 Un CO3 PO1
3. Compare the K-means algorithm with other clustering algorithms such as 2 Un CO3 PO1
hierarchical clustering or DBSCAN.
4. How does clustering around medoids differ from K-means? 2 Un CO3 PO1
5. What is clustering around medoids, and how does it differ from centroid-based 2 Un CO3 PO1
clustering?
6. What is a medoid and how is it defined within a cluster? 2 Un CO3 PO1
7. List the types of hierarchical clustering analysis and draw its dendogram. 2 Re CO3 PO1
8. How does silhouette score help to validate the clusters formed? 2 Un CO3 PO1
9. How PCA can be used for feature extraction and compression? 2 Un CO3 PO2
10. How does ICA differ from Principal Component Analysis (PCA)? 2 Un CO3 PO1
11. What is Factor Analysis (FA) and what problem does it aim to solve? 2 Un CO3 PO1
12. How does the algorithm of ICA work in separating mixed signals? 2 Un CO3 PO1
S.No. PART- B Mar Blooms CO PO
ks level
1 a Explore the challenges and strategies for choosing the optimal number of clusters 8 Un CO3 PO1
(K) in the K-means algorithm.
b Consider a simple dataset with two features, X and Y, and group the data into 8 Ap CO3 PO2
two clusters using the K-means algorithm.
Dataset:
X Y
1 2
2 1
2 3
3 2
6 5
7 4
7 6
8 5

2 a Discuss the role of feature selection and dimensionality reduction techniques in 10 Un CO3 PO1
preprocessing data for K-means clustering. How do these preprocessing steps
influence the clustering results and improve clustering accuracy?
 b Explain the different types of linkages used to cluster the objects with the help of 6 Un CO3 PO1
neat diagram.
3 a Consider a dataset with 5 objects and 2 features (height & weight) for each object. 10 Ap CO3 PO3
Assume k =2 and cluster the objects based on distance metrics. Choose person2
and person 3 are two centroids.

b Compare agglomerative clustering and divisive clustering. 6 Un CO3 PO1

4 a Compute the principal components of the following data: 8 Ap CO3 PO2
CLASS 1
X = 2, 3, 4
Y = 1, 5, 3
CLASS 2
X = 5, 6, 7
Y = 6, 7, 8
b 8 Un CO3 PO2

5 a Consider a database shown below which has 5 transactions. Analyze the database 10 Ap CO3 PO3
given and find the transaction ID under each cluster using k-means algorithm. Use
K=2 and assume A and C are the initial means. Justify your reasoning at each step,
including how the centroids are updated and how the final clusters are formed.

Transaction X1 X2
ID
A 1 1
B 1 0
C 0 2
D 2 4
E 3 5
b Discuss how does the factor analysis model assess the goodness of fit. 6 Un CO3 PO1

6 a Here the feature vectors which depicts the yearly rainfall and runoff rate for 5 years is 10 Ap CO3 PO2
given.

Year 1 2 3 4 5
Rainfall(cm)
10 15 13 90 5

Runoff(cm) 42 46 26 39 29

Determine the principal components by computing the

(i) Covariance matrix
(ii) Eigenvalues and eigenvectors of the covariance matrix

b Illustrate the hierarchy of clustering with the help of dendogram. 6 Un CO3 PO1

UNIT - IV

S.No. PART-A Mar Blooms CO PO

ks level
1. What is Gibbs sampling and what problem does it solve in machine learning? 2 Un CO4 PO1
2. What is hidden Markov model? 2 Re CO4 PO1
3. How does Markov Chain Monte Carlo (MCMC) differ from other optimization 2 Un CO4 PO1
techniques?
4. What are some common types of proposal distributions used in Markov Chain 2 Un CO4 PO1
Monte Carlo algorithms?
5. What is a Bayesian Belief Network (BBN) and how does it relate to probabilistic 2 Un CO4 PO1
 graphical models?
6.How are potential functions used in Markov Random Fields? 2 Un CO4 PO1
7.Can random forest algorithm be used for both continuous and categorical target 2 Un CO4 PO1
variables? Why?
8. What do you mean by ensembling in Machine learning? Is it can be used for 2 Un CO4 PO1
regression?
9. How does Boosting algorithms work? 2 Un CO4 PO1
10. List the three problems of HMM in machine learning. 2 Re CO4 PO1
11. What is the stacking technique in ensemble methods? 2 Un CO4 PO1
12. How do bagging and boosting differ in their approach to improving model 2 Re CO4 PO1
performance?
S.No. PART - B Mar Blooms CO PO
ks level
1 a Consider the following set of training examples shown in Table 10 Ap CO4 PO2

i. What is the entropy of this collection of training examples with respect to

the positive class?
ii. What are the information gains of a1 and a2 relative to these training
examples?
iii. What is the best split (between a1 and a2) according to the classification
error rate?
iv. What is the best split (between a1 and a2) according to the Gini index?
v. What is the information gain of A1 & A2 relative to these training
examples?
b Compare the different boosting algorithms. 6 Un CO4 PO1

2 a The state transition diagram for the flea hops around on a triangle is given. At 10 Ap CO4 PO2
each step it moves to one of the other two vertices at random. What is the
expected time taken for flea to get from vertex 1 to vertex 2?

b Explain the relationship between a Markov Model and a Markov Chain. 6 Un CO4 PO1

3 a Describe how Bayesian Networks can be used to model complex probabilistic 8 Un CO4 PO2
relationships. Provide an example of a real-world application.
b Will the ensemble help to improve the classification accuracy? What are the 8 Un CO4 PO1
benefits of ensemble model? Explain.
4 a Discuss the importance of Markov Chains in the context of Markov Chain Monte 10 Ap CO4 PO2
Carlo (MCMC) methods. How do they ensure proper sampling?
 b Discuss the advantages of using ensemble learning techniques over single models 6 Un CO4 PO1

5 a Analyze the process and importance of tracking methods in sequential data 8 Un CO4 PO3
analysis. Compare and contrast how models like Hidden Markov Models (HMMs)
and Kalman filters are used for state estimation and prediction in dynamic systems.
Provide examples from areas such as signal processing or navigation, and discuss
the strengths and limitations of each method in these applications.
b Explain AdaBoost algorithm covering its key concepts, components, and the 8 Un CO4 PO1
overall algorithmic process.
6 a Describe the stacking method in ensemble learning. Explain how it works, and 8 Un CO4 PO1
provide examples of its application in improving model performance.
b What is bagging in the context of machine learning? How does bagging help to 8 Un CO4 PO1
improve the performance of machine learning models? Explain it in detail.
UNIT - V

S.No. PART - A Mar Blooms CO PO

ks level
1. How reinforcement learning is useful in game playing? 2 Un CO5 PO1
2. Define Q function. 2 Re CO5 PO1
3. What is policy search? 2 Re CO5 PO1
4. Compare active and passive reinforcement learning. 2 Un CO5 PO1
5. What is the difference between episodic and continuing tasks in reinforcement 2 Un CO5 PO1
learning?
6. What is exploration-exploitation tradeoff in reinforcement learning? 2 Un CO5 PO1
7. Define reinforcement learning. Discuss various types of reinforcement learning. 2 Un CO5 PO1
8. Distinguish between model based and model free learning 2 Un CO5 PO2
9. List down at least four different applications for reinforcement learning. 2 Un CO5 PO1
10. Mention significance of "reward function" in reinforcement learning. 2 Un CO5 PO1
11. How do you define a utility function for a given problem in reinforcement 2 Un CO5 PO1
learning?
12. What are some common exploration strategies used in active reinforcement 2 Un CO5 PO1
learning?
S.No. PART - B Mar Blooms CO PO
ks level
a Explain the few common active and passive reinforcement learning techniques. 10 Un CO5 PO1
1 i. Direct Utility Estimation
ii. ADP with exploration function
iii. Q-Learning
b Define the state, action and reward. 6 Re CO5 PO1

2 a Explain the applications and real-world challenges of reinforcement learning in 10 Un CO5 PO1
robotics.
b Write reinforcement learning problem characteristics. 6 Un CO5 PO1

3 a How reinforcement learning problem differs from the other function approximation 10 Un CO5 PO2
tasks? Explain.
b Illustrate the temporal difference learning model. 6 Un CO5 PO1

4 a What is reinforcement learning? Explain the applications and real-world challenges 10 Un CO5 PO1
of reinforcement learning in healthcare applications.
b Discuss the different application, challenges and limitations of policy search in 6 Un CO5 PO1
reinforcement learning.
5 a You are asked to solve the problem of identifying fake bloggers. A fake blogger is 10 Ap CO5 PO3
one who has posted at least 10% of fake blogs. As you are aware, blogs can be in
multiple languages and individual blogs can be in a mix of language. You have
been given the identifiers of nearly 1000 bloggers and you have been told at least
20% of the bloggers post fake blogs. You decided to solve the problem using a
machine learning approach. Answer the following questions in this context:
 1. What are the features that you would consider? Name them and justify their
relevance.
2. What approach do you use to extract these features? Discuss how do you store
them.
3. Determine the appropriate machine learning algorithm and construct a flow
diagram depicting how you would determine the fake bloggers.
4. Discuss the metrics used to verify your algorithm.

(May 2022)
b Describe the concept of discount factor (gamma) in reinforcement learning and its 6 Un CO5 PO1
influence on the agent's behaviour.
6 a List down the difference between reinforcement learning, models based learning 10 Un CO5 PO1
and temporal difference learning. (May
2022)
b Is it necessary to perform generalisation in reinforcement learning? Why? 6 Un CO5 PO1

Course Instructors Course Co-Ordinator Module Co-Ordinator HOD