Machine Learning — DU B. Voc – Software Development Previous Year Question Papers

Your Roll N o............... B.Voc. Software Each question is ol 5 marks. (a) Given the following real-world applications, classify each into Supervised, Unsupervised, or Reinforcement Learning. Justify each in one line. (5) (i) A bank detects fraudulent transactions using past labeled data. (ii) A chatbot improves its responses by receiving a reward when the user gives positive feedback. (iii) A movie streaming app groups viewers based on their watch preferences. (iv) A handwriting recognition model trained with labeled examples of digits. (v) A delivery robot learns the best path to deliver packages by trial and error. (b) A spam email classifier was tested on a dataset of 6000 emails. Out of these, 2000 were spam and 4000 were legitimate. The classifier correctly identified 1700 spam emails, but incorrectly marked 300 legitimate emails as spam. Construct the confusion matrix and compute: (iii) Specificity (c) With the help of an error vs. model complexity diagram. Explain overfitting and underfitting. How do regularization methods improve model generalization and prevent overfitting? (5) (d) What is the importance of feature engineering in ML? Explain any two feature subset selection techniques to deal with high dimensionality problem. (e) Apply the Naive Bayes Classifier to predict whether a car will be stolen or not stolen for the following input features : (5) Color: RED, Type: SUV, Origin: Domestic Use the dataset given below : RED YES SPORTS DOMESTIC SPORTS DON,IESTIC RED YES SPORTS DOMESTIC SPORTS DOMESTIC YELLOW YES SPORTS IMPORTED SW IMPORTED YES YELLOW SUV RED SPORTS IMPORTED YES (f) What are the two popular types of clustering approaches? Describe Hierarchical clustering? (a) Explain why standardization is needed before PCA. How does the PCA algorithm help reduce dimensionality in machine learning? Write the steps of the PCA algorithm. (7) (b) Given the following dataset of weather conditions and whether a person plays tennis, calculate the Information Gain for the attribute "Outlook". Outlook Temperature PlayTennis Sunny Hot No Sunny Mild No Overcast Hot Yes Rainy Cool Yes Rainy Mild Yes Rainy Cool No 3. (a) What type of problems can logistic regression solve? How is it different from linear regression? What is the role of the sigmoid function in logistic regression? (7) (b) Marketing Cosl ({1000) (\) Customers Gained ( ) Use the least squares method to compute the best-fitting regression line. Estimate the number of customers if Rs. 7000 thousand is spent. (8) (a) (i) What is the role of activation functions in a neural network? (ii) Explain why non-linear activation functions are essential in multi-layer networks. (iii) Give examples of commonly used activation functions. (7) (b) How does hold out validation differ from k-fold cross validation? For k- 1 0 and datapoints- D I