Register
Number
SRM Institute of Science and Technology
Set - A
College of Engineering and Technology
School of Computing
SRM Nagar, Kattankulathur – 603203, Chengalpattu District, Tamil Nadu
Academic Year: 2024-25 (odd)
Test: CT I Date: 26-08-2025
Course Code & Title: 21CSE451T & Pattern Recognition Techniques Duration: Two periods
Year& Sem: IV Year /VII Sem [Link]
Course Articulation Matrix:
Course PO1 PO2 PO3 PO4 PO5 PO6 PO7 PO8 PO9 PO10 PO11 PO12
Outcome
CO3 - - - - 1 - - - - - - -
CO4 - - - - 1 - - - - - - -
CO5 - - - - 1 - - - - - - -
Note: CO3 – To identify data manipulation and cleaning techniques using pandas
CO4 – To constructs the Graphs and plots to represent the data using python packages
CO5 – To apply the principles of the data science techniques to predict and forecast the outcome of real-
world problem
Part – A (10 x 1 = 10 Marks)
Instructions:
1) Answer ALL questions.
2) The duration for answering Part A is 15 minutes (this sheet will be collected after 15 minutes).
3) Encircle the correct answer.
Question Marks BL CO PO PI
[Link]
Code
1 A dataset contains Amazon Prime movie reviews marked as positive, 1 1 2 1
negative, or neutral. Which type of machine learning is most appropriate for
building a model that classifies new reviews?
a. Supervised Learning
b. Unsupervised Learning
c. Reinforcement Learning
d. Semi-supervised Learning
Or
The first step in the design cycle of a pattern recognition system is:
a. Problem definition
b. Feature extraction
c. Classifier selection
d. Evaluation
2 In Bayesian estimation, the posterior is proportional to: 1 1 2 1
a. Prior / Likelihood
b. Likelihood × Prior
c. Likelihood + Prior
d. Evidence × Prior
3 Two Gaussian distributions have: 1 2 1 1
p(x|ω1) = N(x; 1, 1)
p(x|ω2) = N(x; 3, 1)
with equal priors.
Find the decision boundary.
a. x = 2
b. x = 1
c. x = 1.5
d. x = 2.5
4 Dangerous fires are rare and let it be 1%. But smoke is common, and it is 1 2 1 1 1.1.3
10%. Also 90% of dangerous fires make smoke (probability of smoke with
respect to fire). Estimate the probability of dangerous fire when there is
smoke using Bayes’ Rule.
a. 0.01
b. 0.10
c. 0.09
d. 0.90
5 In the context of machine perception, which of the following best describes 1 2 3 1
the challenge known as the "inverse problem"?
a. Converting sensory signals into motor commands
b. Deriving high-level decisions from low-level data
c. Inferring the underlying causes of sensory input from observed
data
d. Mapping output features back to the original data domain
6 The goal of Bayesian parameter estimation is to: 1 1 4 2
a. Minimization of variance
b. Computation of posterior parameters
c. Avoid priors
d. Maximization of training accuracy
of data D. The prior distribution is θ ∼ N(0, 1), and the likelihood P(D | θ) is
7 Consider a scenario where a parameter θ is being estimated using a sample 1 1 4 2
maximized at θ̂ = 5. Which of the following statements is most accurate?
a. The Maximum Likelihood Estimate (MLE) of θ will be closer to 0
than the Bayesian Parameter Estimate (MAP).
b. The MAP estimate of θ will be exactly 5, same as MLE.
c. The MAP estimate of θ will lie between 0 and 5 due to the
influence of the prior.
d. The MLE will be a random variable due to the prior.
8 The curse of dimensionality refers to: 1 1 4 2
a. Memory overload
b. Infinite decision surfaces
c. Sparsity of data in high dimensions
d. Overtraining of low-dim data
9 Linear discriminant functions in high-dimensional space: 1 2 5 2
a. Always produce circular decision boundaries
b. Become unstable due to data sparsity
c. Increase model interpretability
d. Ignore covariance
10 For a Hidden Markov Model (HMM), which method is typically used to 1 2 5 2
estimate the most probable sequence of hidden states given the observations?
a. Baum-Welch Algorithm
b. Viterbi Algorithm
c. Forward Algorithm
d. Backward Algorithm
Register
Number
SRM Institute of Science and Technology
Set -
College of Engineering and Technology
School of Computing
SRM Nagar, Kattankulathur – 603203, Chengalpattu District, Tamil Nadu
Academic Year: 2024-25 (EVEN SEM)
Test: FT4 Date:29-04-2025
Course Code & Title: 21CSS303T-Data Science Duration: Two periods
Year& Sem: III Year /VI Sem [Link]
Part – B (4 x 5 = 20 Marks)
Instructions: Answer ALL Questions
Q. Question Marks BL CO PO PI
No Code
11 Distinguish between Supervised and Unsupervised Learning with 5 2 2 1 1.1.1
Examples.
Ans.
Aspect Supervised Learning Unsupervised Learning
Learns from labeled
Learns from unlabeled
Definition data (input-output
data (only inputs)
pairs)
Predict outcomes for Discover hidden patterns
Goal
new inputs or structure in data
Clustering,
Classification,
Examples Dimensionality
Regression
Reduction
Data Requires large labeled Works with unlabeled
Requirement datasets data
Spam detection, Customer segmentation,
Use Cases
disease diagnosis market basket analysis
12 In a particular binary hypothesis testing application, the conditional 5 2 1 1 1.1.3
density for a scalar feature x given class w₁ is
p(x | w₁) = k₁ * exp(-x² / 20)
Given class w₂, the conditional density is
p(x | w₂) = k₂ * exp(-(x - 6)² / 12)
a. Find the values of k₁ and k₂ such that the conditional densities
are valid probability density functions.
b. Assume that the prior probabilities of the two classes are equal,
and that the cost for choosing correctly is zero. If the costs for
choosing incorrectly are C12=√ 3 and C21=√ 5 . What is the
expression for the conditional risk?
Ans.
a. We solve for the parameters k₁ and k₂ by recognizing that the two
equations are in the form of the normal Gaussian distribution:
k₁ * e^(−x²/20) = (1 / sqrt(2πσ₁²)) * e^(−(x−μ₁)² / (2σ₁²))
By comparing exponents:
−x²/20 = −(x − μ₁)² / (2σ₁²)
Let μ₁ = 0 and σ₁² = 10
Then,
k₁ = 1 / √ 2 π∗10
Similarly, for the second Gaussian:
μ₂ = 6, σ₂² = 12
k₂ = 1 / √ 2 π∗12
b. The conditional risk for two-category classification is given by:
R(α₁ | x) = λ₁₁ * P(ω₁ | x) + λ₁₂ * P(ω₂ | x)
R(α₂ | x) = λ₂₁ * P(ω₁ | x) + λ₂₂ * P(ω₂ | x)
Given: Prior probabilities of the two classes are equal:
P(ω₁) = P(ω₂) = 1/2
Costs for decisions:
λ₁₁ = 0, λ₂₂ = 0 (correct decisions have zero cost)
λ₁₂ = C₁₂ = √3
λ₂₁ = C₂₁ = √5
Thus, the expression for conditional risk when choosing class ω₁ (α₁):
R(α₁ | x) = λ₁₂ * P(ω₂ | x) = √3 * P(ω₂ | x)
Similarly, the expression for conditional risk when choosing class ω₂
(α₂):
R(α₂ | x) = λ₂₁ * P(ω₁ | x) = √5 * P(ω₁ | x)
13 Explain Bayesian Decision Theory in detail. 5 2 2 2 2.1.3
Ans.
Bayesian Decision Theory is a fundamental statistical approach to the
problem of pattern classification. It provides a theoretical framework for
decision making under uncertainty. It is considered optimal because it
minimizes the probability of misclassification, assuming that the true
distributions of the data are known.
Bayes’ theorem provides a way to update the probability estimate for a
hypothesis as more evidence or information becomes available:
P(ωi|x) = [P(x|ωi)P(ωi)] / P(x)
Where:
- P(ωi|x): Posterior probability of class ωi given observation x
- P(x|ωi): Likelihood of observation x under class ωi
- P(ωi): Prior probability of class ωi
- P(x): Evidence or total probability of x across all classes
Bayesian Decision Rule: The Bayes decision rule is to choose the class
that has the highest posterior probability. That is, decide ωi if P(ωi|x) >
P(ωj|x) for all j ≠ i.
In the case of 0-1 loss function (equal cost for all types of errors), this
rule minimizes the probability of error.
Minimum Risk Classification: If different types of classification errors
have different costs, the decision should minimize the expected risk:
R(αi|x) = ∑ λ(αi|ωj) P(ωj|x)
Where:
- R(αi|x) is the expected risk for choosing action αi
- λ(αi|ωj) is the loss incurred for choosing αi when the true class is ωj
- The decision rule becomes: Choose the action αi that minimizes R(αi|x)
14 Explain the need for dimensionality reduction and mention two popular 5 3 2 2 2.2.3
techniques.
Or
Explain the curse of dimensionality. How do component analysis
techniques like PCA help in dealing with it?
Ans.
In machine learning and pattern recognition, high-dimensional data often
leads to challenges such as overfitting, increased computation, and poor
generalization. This phenomenon is referred to as the 'curse of
dimensionality'. As the number of features increases, the volume of the
feature space grows exponentially, making data sparse and less
informative for training models.
Dimensionality reduction helps by transforming high-dimensional data
into a lower-dimensional space while retaining as much relevant
information as possible. It improves computational efficiency, reduces
storage requirements, helps in visualization, and enhances model
performance by eliminating noise and redundant features.
Two popular dimensionality reduction techniques are:
• Principal Component Analysis (PCA): An unsupervised
technique that projects the data onto new axes (principal components)
that maximize variance. It helps in capturing the most important patterns
in the data with fewer dimensions. (Explain it)
• Linear Discriminant Analysis (LDA): A supervised method that
finds a projection maximizing the separation between multiple classes. It
is especially useful for classification problems. (Explain it)
Part – C (2 x 10 = 20 Marks)
Instructions: Answer ALL questions.
Q. Question Marks BL CO PO PI
No Code
15 a Explain the concept of likelihood ratio and minimum-error-rate 10 2 3 5
classification. Derive the decision rule using Bayesian decision theory
for two-class classification.
Ans.
Liklihood ratio:
Minimum error rate classification:
Bayesian Decesion Theory:
(OR)
15 b Discuss the role of discriminant functions in pattern recognition. 10 1 3 1
Derive the discriminant function for normally distributed classes with
equal covariance matrices.
Ans.
Case:
16 a Discuss Maximum Likelihood Estimation (MLE). Derive the MLE for 10 2 2 2 2.2.3
the mean of a Gaussian distribution with known variance.
Ans.
(OR)
16 b Derive the Bayesian parameter estimation for the mean of a Gaussian 10 2 2 2 2.2.3
distribution with known variance and a Gaussian prior.
Ans.
Course Outcome (CO) and Bloom’s level (BL) Coverage in Questions
CO Coverage
60% 55%
50% 45%
40%
30%
20%
10%
0%