AKTU Machine Learning Techniques (KCS055) Question Paper 2023-24
| University | Dr. A.P.J. Abdul Kalam Technical University (AKTU), Lucknow |
|---|---|
| Course | B.Tech, Semester V |
| Subject | Machine Learning Techniques (KCS055) |
| Exam | Theory Examination 2023-24 (odd semester, held January 2024) |
| Time · Marks | 3 hours · 100 marks |
| Source | AKTU question paper, via Notes Gallery |
Note: Attempt all sections. If any data is missing, choose suitably.
Section A
Q1. Attempt all questions in brief · 2 × 10 = 20 marks
- Discuss the important objectives of Machine Learning. [2 · CO1]
- Discuss overfitting and underfitting situation in decision tree learning. [2 · CO1]
- Discuss support vectors in SVM. [2 · CO2]
- What is gradient descent delta rule? [2 · CO2]
- Explain Case-based learning. [2 · CO3]
- For which problem decision tree is best suitable. [2 · CO3]
- Define the term ANN, and CNN. [2 · CO4]
- Differentiate between Lazy and Eager Learning. [2 · CO4]
- Comparison of purely analytical and purely inductive learning. [2 · CO5]
- Define the term Offspring, Chromosome and Genes are used in GA. [2 · CO5]
Section B
Q2. Attempt any three · 10 × 3 = 30 marks
- Compare Supervised and Unsupervised Learning Techniques with examples. [10 · CO1]
- Explain Maximum Likelihood and Least Squared Error Hypothesis with example. [10 · CO2]
- Compare and contrast Information Gain, Gain Ratio, and Gini Index in detail. [10 · CO3]
- Explain the different layers used in convolutional neural network with suitable examples. [10 · CO4]
- Discuss the applications of reinforcement learning. In which problems reinforcement learning is used? [10 · CO5]
Section C
Q3 to Q7. Attempt any one part of each question · 10 × 5 = 50 marks
Q3
- Compare regression, classification and clustering in machine learning along with suitable real life applications. [10 · CO1]
- Explain the “Concept Learning” Task Giving an example. [10 · CO1]
Q4
- Explain hyperplane (decision boundary) in SVM. Categorize various popular kernels associated with SVM. [10 · CO2]
- Differentiate between Naïve Bayes classifier and Bayesian belief networks. Give an application of Bayesian belief networks. [10 · CO2]
Q5
- Discuss Decision Tree and explain its working in detail. [10 · CO3]
- Demonstrate K-Nearest Neighbors algorithm for classification with the help of an example. [10 · CO3]
Q6
- Illustrate backpropagation algorithm by assuming the training rules for output unit weights and Hidden Unit weights. [10 · CO4]
- Write short notes on Probably Approximately Correct (PAC) learning model. [10 · CO4]
Q7
- Explain Q-learning with its key terms, key feature and elements. Discuss its applications used in real life. [10 · CO5]
- Define the term Genetic Algorithm. Discuss the working of Genetic algorithm with the help of flowchart. [10 · CO5]
Most repeated topics
How many of the last 6 AKTU Machine Learning Techniques papers (KCS055/BCS055, 2020-21 to 2025-26) asked about each topic. Prepare the top ones first.
Counted by LearnAIFast from the 2020-21, 2021-22, 2022-23, 2023-24, 2024-25 and 2025-26 odd-semester theory papers.
Answer hints
Q1(c). Support vectors in SVM
Support vectors are the training points that lie closest to the decision boundary (hyperplane), on the edges of the margin. They alone decide where the hyperplane sits: remove any other point and the boundary doesn't change. SVM chooses the hyperplane that maximises the margin, the distance to these support vectors.
Q1(d). Gradient descent delta rule
The delta rule trains a linear unit by reducing the squared error E = ½ Σ (t − o)². Each weight moves a small step against the error gradient: Δwᵢ = η Σ (t − o) xᵢ, where η is the learning rate, t the target and o the output. Unlike the perceptron rule, it still converges to the best-fit weights when the data is not linearly separable.
Q1(h). Lazy vs eager learning
Lazy learners (KNN, case-based reasoning, locally weighted regression) store the training data and do the work only when a query arrives, so training is fast and prediction is slow. Eager learners (decision trees, neural networks, SVM, Naive Bayes) build a general model during training, so prediction is fast but the model is fixed in advance.
Q1(j). Offspring, chromosome and genes in GA
A chromosome is one candidate solution, usually written as a string such as 101101. Genes are the individual positions in that string, each encoding one part of the solution. Offspring are the new chromosomes created from selected parents through crossover and mutation, forming the next generation.
Q2(c). Information gain vs gain ratio vs Gini index
Information gain (used by ID3) is the drop in entropy after a split, but it favours attributes with many values. Gain ratio (used by C4.5) divides gain by the split information to correct that bias. Gini index (used by CART) measures impurity as 1 − Σ pᵢ² and is cheaper to compute because it has no logarithm. In an answer, give each formula, one small worked example, and which algorithm uses it.
Q7(a). Q-learning key terms
Cover agent, environment, state, action, reward, policy and the Q-table. The update rule is Q(s, a) ← Q(s, a) + α [r + γ · max Q(s′, a′) − Q(s, a)], where α is the learning rate and γ the discount factor. Key features: it is model-free and off-policy, and it balances exploration and exploitation (for example ε-greedy). Applications include game playing, robot navigation, traffic signal control and recommendation systems.
Use our prompt guide to have AI quiz you on any topic above. Always check AI answers against your notes.
Other years
This paper is shared for study purposes only. Copyright belongs to AKTU. If you own the copyright and want it removed, email learnwithskillnest@gmail.com and we will take it down.