c) Draw a fully connected and labeled Neural Network diagram with the following specifications: 3 Inputs 3 Hidden Layers An output [8] d) Write Python code that encodes the Color column in the left pane so that it appears as shown in the right pane. Color 0 Red 0 1 1 Green 2 Blue QUESTION 4 Refer to the diagram below for a iris dataset stored in a dataframe called df sepal-length sepal-width petal-length petal-width 2 count 150.00800 150.000000 150.009006 5 843333 3 054090 150. 800800 3.758667 1.198667 6. 828065 0.433594 1.764420 0763161 4.3080 2-60806 2-86 1 6060 6.30 75% 1.380 max a) What is a dataframe? [2] b) Write the python code that produced this table [2] c) Define the term skewness [4] d) Comment, with justification, the skewness of the petal-length column. (2] e) Explain with the aid of a labelled diagram, the concept of gradient descent [10] QUESTION 5 You work at a social media company, and your task is to detect cyberbullying messages based on the text they contain. You have access to a large number of messages, N, which have been manually labelled as "OK" and "bullying". a) Describe how can you apply a logistic regression classifier to the task and evaluate it. [8) b) You decide to use precision and recall instead of accuracy as the evaluation metric for this task. Why does this decision make sense? page 3 of 4 SECTION B Answer any THREE questions in this section QUESTION 2 You are tasked with building a model to predict house prices in Masvingo city based on various features (e.g., size in square km, number of bedrooms) a) Explain the steps you would take to prepare the data for model training in; i. data cleaning ii. transformation, and ifi, preprocessing. 131 b) Explain why a regression approach might be suitable for this task. c) Describe the potential challenges you might encounter when building a regression model for this problem. (4] d) Describe how you would evaluate the effectiveness of your final model. 141 QUESTION 3 Imagine you are working on a spam email classification problem. a) Explain why a binary classification model might be appropriate for this task. [2] b) Discuss the advantages and disadvantages of using logistic regression and decision trees for this specific problem. c) Explain how you would choose the most suitable model for this task? [2] Regularization is a key concept to prevent overfitting in machine learning models d) Explain the concept of overfit ing and how it can negatively impact model performance. 141 e) Compare and contrast how LI and L2 regularization techniques work and how they help to ad tress overfitting. Page 3 of 4 QUESTION 4 Both gradient boosting and AdaBoosting are ensemble learning techniques that leverage decision trees a) Explain the core concept behind ensemble learning methods. [2] b) Compare and contrast gradient boosting and AdaBoosting algorithms in terms of their working principles and how they address limitations of single decision trees. Random forests are popular ensemble methods known for their robustness and accuracy. ) Explain how the concept of feature randomness is used in building a random forest model. 12] d) Discuss the advantages of using random forests for complex machine learning problems. [6] e) How can you assess the feature importance within a random forest model? [4] QUESTION 5 You are given a dataset containing customer purchase history. a) Explain how you would analyse this data using clustering techniques. [4] b) Discuss the potential challenges associated with choosing the optimal number of clusters (K) in K-means clustering. [5) Neural networks are powerful tools for complex pattern recognition tasks. ) Describe the basic architecture of a feedforward neural network and explain how information propagates through the network. 15) d) Choose a specific application (e.g., image recognition, speech recognition) and discuss how neural networks can be used to address this task. [6] ==== END OF PAPER = === Page 4 of 4