UNIT-3
1. Density Estimation
Q1. The main disadvantage of histogram-based density estimation is:
A. It is too fast
B. It requires a parametric form
C. It is sensitive to bin width and origin
D. It uses kernel functions
Answer: C
2. Parzen Windows
Q2. As the window width (h) in Parzen estimation becomes very small, the resulting estimate
becomes:
A. Smoother
B. More biased
C. More discrete and noisy
D. Less sensitive to outliers
Answer: C
3. K-Nearest Neighbor Estimation
Q3. In K-NN density estimation, the density is estimated by:
A. Counting total samples
B. Averaging distances of all points
C. Volume enclosing K neighbors
D. Distance to the farthest class centroid
Answer: C
4. The Nearest Neighbor Rule
Q4. The asymptotic error rate of 1-NN classifier is:
A. Equal to Bayes error
B. Greater than Bayes error but less than twice it
C. Always zero
D. Equal to training error
Answer: B
5. Metrics and Nearest Neighbor Classification
Q5. Mahalanobis distance is preferred over Euclidean distance when:
A. Features are binary
B. Features are independent
C. Features are correlated
D. Features are normalized
Answer: C
6. Fuzzy Classification
Q6. In fuzzy K-NN, a sample is assigned to:
A. Only the closest class
B. All classes with crisp probability
C. One class with 100% certainty
D. Multiple classes with varying membership values
Answer: D
7. Reduced Coulomb Energy (RCE) Networks
Q7. In an RCE network, the radius of a receptive field is adjusted based on:
A. Number of neighbors
B. Activation function
C. Classification error
D. Distance to training points and threshold
Answer: D
8. Approximations by Series Expansions
Q8. In Fourier series expansion, increasing the number of terms results in:
A. Lower training accuracy
B. Smoother approximations
C. More random outputs
D. Less generalization
Answer: B
UNIT -4
Linear Discriminant
Linear Discriminant Functions and Decision Surfaces
Q1. A linear discriminant function in two dimensions represents a:
A. Point
B. Line
C. Curve
D. Hyperplane
Answer: B
Generalized Linear Discriminant Functions
Q2. A generalized linear discriminant function can be used to:
A. Classify only two categories
B. Handle nonlinear separability
C. Solve multi-class problems
D. Estimate probabilities directly
✅ Answer: C
Two-Category Linearly Separable Case
Q3. In a two-class linearly separable problem, the decision boundary is:
A. Non-existent
B. A quadratic curve
C. A straight line
D. A cluster center
Answer: C
Perceptron Criterion Functions
Q4. The perceptron criterion function minimizes:
A. Total distance between classes
B. Misclassified samples' distance from the boundary
C. Weight norm
D. Variance within a class
✅ Answer: B
Relaxation Procedures
Q5. Relaxation procedures in pattern classification are used to:
A. Speed up convergence
B. Reduce noise
C. Minimize energy functions
D. Ensure orthogonality
Answer: A
Nonseparable Behavior
Q6. When data is not linearly separable, the perceptron:
A. Always converges
B. Gives infinite solutions
C. Does not converge
D. Classifies all correctly
Answer: C
Minimum Squared Error Procedures
Q7. Minimum squared error methods are equivalent to:
A. Linear regression
B. Perceptron
C. Maximum likelihood
D. Histogram equalization
Answer: A
Linear Programming Algorithms
Q8. Linear programming methods in classification are mainly used to:
A. Find kernel functions
B. Solve optimization problems with linear constraints
C. Reduce dimensionality
D. Calculate distances
Answer: B
🔹 Support Vector Machines
Q9. The support vectors in an SVM are:
A. Points far from the decision boundary
B. Outliers
C. Points lying on the margin
D. Random samples
Answer: C
Unit-5
ONE MARK QUESTIONS (Intermediate Level MCQs)
Introduction to Neural Networks
Q1. The function of an activation function in a neural network is to:
A. Multiply inputs by weights
B. Perform output normalization
C. Introduce non-linearity
D. Store weights
Answer: C
Multilayer Neural Networks: Feedforward & Classification
Q2. In a feedforward neural network, data flows:
A. In a loop
B. Bidirectionally
C. In one direction from input to output
D. Randomly between layers
Answer: C
Backpropagation Algorithm
Q3. Backpropagation minimizes the error by adjusting:
A. Output values
B. Learning rate
C. Weights using gradient descent
D. Bias only
Answer: C
Decision Trees
Q4. In a decision tree, the attribute chosen for a split is based on:
A. Variance
B. Information gain or Gini index
C. Mean square error
D. Learning rate
Answer: B
CART (Classification and Regression Trees)
Q5. The CART algorithm generates:
A. Only classification rules
B. Only regression outputs
C. Binary decision trees
D. Fuzzy inference rules
Answer: C
Applications: Face Recognition System
Q6. In face recognition, neural networks are primarily used for:
A. Data compression
B. Pattern classification
C. Language translation
D. Hashing images
Answer: B
