Makaut question 2024

GROUP A
Group A: Very Short Answer Type Questions
Q1. For a linear classifier g(x)=wTx+w0g(x)=wTx+w0, what is the angle
between ww and a vector lying on the hyperplane?
 Explanation:
A hyperplane is defined by the equation wTx+w0=0wTx+w0=0.
The vector ww is always perpendicular (normal) to the hyperplane.
Therefore, the angle between ww and any vector lying on the
hyperplane is 90 degrees (or π/2π/2 radians).
Answer: 90 degrees (or π/2π/2 radians).

Q2. What is a decision tree?

 Explanation:
A decision tree is a tree-like structure that is used to make decisions
based on input data.
It consists of nodes (where decisions are made), branches (possible
outcomes), and leaves (final decisions or classifications).
Answer: A decision tree is a flowchart-like structure used for making decisions,
with nodes for tests and leaves for outcomes.

Q3. Mention which tree structure is used to represent the process of

hierarchical clustering.
 Explanation:
In hierarchical clustering, a tree called a dendrogram is used to represent
how clusters are merged or divided at different levels.
Answer: Dendrogram.

Q4. What is feature extraction?

 Explanation:
Feature extraction involves transforming raw data into a set
of important features (attributes) that are useful for analysis or
classification.
Answer: It’s the process of transforming raw data into meaningful features for
analysis or classification.

Q5. Underfitting occurs when the model is not able to obtain a sufficiently
low error value on the ____.
 Explanation:
Underfitting means that the model is too simple and does not fit the
training data well.
This happens when the model doesn’t fit the training data.
Answer: training data.

Q6. When do we use maximum likelihood estimation (MLE)?

 Explanation:
MLE is used to find the parameters of a statistical model that make the
observed data most likely.
It’s commonly used in statistics and machine learning for parameter
estimation.
Answer: When estimating parameters that best explain the observed data.

Q7. Is an HMM (Hidden Markov Model) a finite-state machine?

 Explanation:
An HMM is a probabilistic finite-state machine because it has a finite
number of states and transitions between them based on probabilities.
Answer: Yes.

Q8. What are the purposes of dimensionality reduction?

 Explanation:
 Reduce data size and complexity.
 Remove noise.
 Make models faster and more accurate.
 Help visualize high-dimensional data.
Answer: To reduce data size and noise, improve model performance, and aid
visualization.

Q9. What is the non-parametric method of estimation?

 Explanation:
Non-parametric methods do not assume a specific form for the data
distribution.
Examples: k-nearest neighbors (KNN), kernel density estimation.
Answer: Methods that do not assume a specific data distribution, e.g., KNN.

Q10. In the perceptron criterion, what is the expression for sequential

update of the weight vector?
 Explanation:
For a misclassified sample (xi,yi)(xi,yi),
Update rule:
wnew=wold+ηyixiwnew=wold+ηyixi
where ηη is the learning rate.
Answer: wnew=wold+ηyixiwnew=wold+ηyixi.

Q11. What is the acronym of CART?

 Explanation:
CART stands for Classification And Regression Trees.
Answer: Classification And Regression Trees.

Q12. The k-means algorithm is sensitive to ____.

 Explanation:
K-means is sensitive to the initial choice of cluster centers (starting
points).
Answer: Initial cluster centers.
GROUP B
Q2. Define pattern, feature, feature vector, feature space, and feature
extraction.
✅ Pattern:
A pattern is an object, event, or phenomenon that can be recognized or
classified.
For example, a handwritten digit “7” or an apple image.
✅ Feature:
A feature is a measurable property or attribute of a pattern that helps in
distinguishing it from others.
Example: In image recognition, features can be color, shape, edges, or
textures.
✅ Feature Vector:
A feature vector is a collection of numerical values (features) that represent a
pattern in a structured form.
Example: A grayscale image could have pixel intensity values as its feature
vector.
✅ Feature Space:
The multi-dimensional space in which feature vectors reside. Each axis
represents a feature.
Example: If you have 3 features, the feature space is 3D.
✅ Feature Extraction:
The process of transforming raw data into feature vectors by selecting the
most informative features.
Example: Converting a raw image into edge or corner features.
🔎 Example (for clarity):
Let’s say we want to classify fruits based on color and size.
 Pattern: An apple
 Features: Color (red), Size (medium)
 Feature Vector: (red, medium)
 Feature Space: 2D space (color axis and size axis)
 Feature Extraction: Converting an image to numeric color/size values

Q6. Explain the various steps used in pattern recognition systems.

✅ Steps in a Pattern Recognition System:
1. Data Acquisition
 Collect data (e.g., images, audio).
 Example: Taking a photo of a handwritten digit.
2. Preprocessing
 Clean and normalize data.
 Example: Remove noise or resize images.
3. Feature Extraction
 Extract relevant features from raw data.
 Example: Find edges in an image.
4. Feature Selection / Dimensionality Reduction
 Choose the most useful features.
 Example: Using PCA to reduce dimensions.
5. Classification / Decision Making
 Use models (e.g., SVM, k-NN) to classify data into categories.
 Example: Recognize digit “7” from an image.
6. Post-processing / Evaluation
 Refine results and check accuracy.
 Example: Evaluate model performance on test data.
🔎 Example:
 Input: Image of digit 7
 Preprocessing: Grayscale conversion
 Feature extraction: Pixel intensities
 Feature selection: PCA to reduce data
 Classification: SVM to classify digit
 Evaluation: Check accuracy on test images
GROUP C
Show that the weight vector of a linear classifier g(x)=wTx+w0 is normal to
the vectors xi and xj lying on the hyperplane.
Step 1: Hyperplane definition
The hyperplane is defined as:
wTx+w0=0
Step 2: Two points on the hyperplane
Let:
xi and xj
be two distinct points on this hyperplane. Because they lie on the hyperplane:
wTxi+w0=0
wTxj+w0=0

Step 3: Subtract to find the directional vector in the hyperplane

Subtract the second equation from the first:
(wTxi+w0)−(wTxj+w0)=0
wT(xi−xj)=0
Let:
d=xi−xj

Step 4: Interpretation
The result:
wTd=0
means that:
w⋅d=0

Conclusion
Since the dot product of w and d is zero, w is orthogonal (normal) to
any difference vector d=xi−xj within the hyperplane.
Therefore, the weight vector ww is normal (perpendicular) to the
hyperplane defined by the linear classifier.

Two-class problem (binary classification)

A two-class problem — also called a binary classification problem —
involves separating data points into two distinct categories.

Key characteristics:
✅ Two classes
 Examples:
 Spam or not spam
 Apple or Orange
 Disease positive or negative
 True or False
 Approved or Denied
✅ Labeling
 Labels are typically ±1 or 0,1
 E.g.,y=+1 for positive class,y=−1 for negative class.
✅ Goal
 Train a model (like SVM, logistic regression, decision tree, etc.) to
predict the class of new, unseen data.

Example
Let’s say you’re building an email spam filter:
 Class 1 (+1): Spam
 Class 2 (-1): Not spam
You train the model on examples of spam and not-spam emails, so it can learn a
decision boundary to separate them.
Applications of SVM
1. Text Classification
 Spam detection in emails
 Sentiment analysis (positive/negative classification)
 Document categorization
2. Image Classification
 Face detection
 Handwritten digit recognition (e.g., MNIST dataset)
 Object recognition
3. Bioinformatics
 Protein classification
 Gene expression data analysis
 Cancer classification based on microarray data
4. Speech Recognition
 Speaker identification
 Phoneme classification
5. Financial Applications
 Credit risk analysis
 Stock market prediction
6. Other areas
 Fraud detection
 Time series classification
 Customer churn prediction

Benefits of SVM
1. Effective in High-Dimensional Spaces
  Works well even when the number of features is greater than the
number of samples.
2. Memory Efficient
 Uses only a subset of training points (support vectors) in the
decision function.
3. Versatile with Different Kernels
 Can handle linear and nonlinear classification using kernel trick
(RBF, polynomial, sigmoid kernels).
4. Robust to Overfitting
 Particularly effective in cases where there is a clear margin of
separation.
5. Strong Theoretical Foundation
 Based on the concept of maximizing the margin, which tends to
improve generalization.
6. Works Well with Clear Margin of Separation
 If classes are well-separated, SVM usually performs very well.