Fundamentals of Machine Learning (4341603) - Winter 2023 Solution

Question 1(a) [3 marks]

Define human learning and explain how machine learning is different from human learning?

Answer:

Table: Human Learning vs Machine Learning

• Human Learning: Process of acquiring knowledge through experience, observation, and reasoning
• Machine Learning: Automated learning from data using algorithms to identify patterns

Mnemonic: "Humans Experience, Machines Analyze Data" (HEMAD)

---

Question 1(b) [4 marks]

Describe the use of machine learning in finance and banking.

Answer:

Applications in Finance and Banking:

• Risk Assessment: ML analyzes customer data to predict creditworthiness
• Customer Service: Chatbots provide 24/7 support using NLP
• Regulatory Compliance: Automated monitoring for suspicious activities

Mnemonic: "Finance Needs Smart Analysis" (FNSA)

---

Question 1(c) [7 marks]

Give difference between Supervised Learning, Unsupervised Learning and Reinforcement Learning.

Answer:

Comparison Table:

Key Characteristics:

• Supervised Learning: Teacher-guided learning with correct answers provided
• Unsupervised Learning: Self-discovery of hidden patterns in data
• Reinforcement Learning: Learning through trial and error with rewards/penalties

Mnemonic: "Supervised Teachers, Unsupervised Explores, Reinforcement Rewards" (STUER)

---

Question 1(c OR) [7 marks]

Explain different tools and technology used in machine learning.

Answer:

ML Tools and Technologies:

Key Technologies:

• Cloud Platforms: AWS, Google Cloud for scalable computing
• Development Environments: Jupyter Notebook, Google Colab
• Big Data Tools: Spark, Hadoop for large datasets

Mnemonic: "Python Libraries Visualize Data Effectively" (PLVDE)

---

Question 2(a) [3 marks]

Define outliers with one example.

Answer:

Definition: Outliers are data points that significantly differ from other observations in a dataset.

Example Table:

• Detection: Values beyond 1.5 × IQR from quartiles
• Impact: Can skew statistical analysis and model performance

Mnemonic: "Outliers Stand Apart" (OSA)

---

Question 2(b) [4 marks]

Explain regression steps in detail.

Answer:

Regression Process Steps:

Detailed Steps:

• Data Collection: Gather relevant dataset with input-output pairs
• Preprocessing: Clean data, handle missing values, normalize features
• Feature Selection: Choose relevant variables that affect target
• Model Training: Fit regression line to minimize prediction errors

Mnemonic: "Data Preprocessing Features Train Evaluation Predicts" (DPFTEP)

---

Question 2(c) [7 marks]

Define Accuracy and for the following binary classifier's confusion matrix, find the various measurement parameters like 1. Accuracy 2. Precision.

Answer:

Confusion Matrix Analysis:

Calculations:

Definitions:

• Accuracy: Proportion of correct predictions out of total predictions
• Precision: Proportion of true positive predictions out of all positive predictions

Mnemonic: "Accuracy Counts All, Precision Picks Positives" (ACAPP)

---

Question 2(a OR) [3 marks]

Identify basic steps of feature subset selection.

Answer:

Feature Subset Selection Steps:

Basic Steps:

• Generation: Create different combinations of features
• Evaluation: Test each subset using performance metrics
• Selection: Choose optimal subset based on criteria

Mnemonic: "Generate, Evaluate, Select" (GES)

---

Question 2(b OR) [4 marks]

Discuss the strength and weakness of the KNN algorithm.

Answer:

KNN Algorithm Analysis:

Key Points:

• Lazy Learning: No explicit training phase required
• Distance-Based: Classification based on neighbor proximity
• Memory-Intensive: Stores entire training dataset

Mnemonic: "Simple but Slow, Effective but Expensive" (SBSEBE)

---

Question 2(c OR) [7 marks]

Define Error-rate and for the following binary classifier's confusion matrix, find the various measurement parameters like 1. Error value 2. Recall.

Answer:

Confusion Matrix Analysis:

Calculations:

Definitions:

• Error Rate: Proportion of incorrect predictions out of total predictions
• Recall: Proportion of actual positives correctly identified

Mnemonic: "Error Excludes, Recall Retrieves" (EERR)

---

Question 3(a) [3 marks]

Give any three examples of unsupervised learning.

Answer:

Unsupervised Learning Examples:

• Market Basket Analysis: Finding product purchase patterns
• Social Network Analysis: Identifying community structures
• Anomaly Detection: Detecting unusual patterns in data

Mnemonic: "Customers, Documents, Genes Group Automatically" (CDGGA)

---

Question 3(b) [4 marks]

Find Mean and Median for the following data: 4,6,7,8,9,12,14,15,20

Answer:

Statistical Calculations:

Step-by-step:

• Data: Already sorted: 4,6,7,8,9,12,14,15,20
• Mean: Sum all values ÷ count = 95 ÷ 9 = 10.56
• Median: Middle value in sorted list = 9 (5th position)

Mnemonic: "Mean Averages All, Median Middle Value" (MAAMV)

---

Question 3(c) [7 marks]

Describe k-fold cross validation method in detail.

Answer:

K-Fold Cross Validation Process:

Process Steps:

Advantages:

• Unbiased Estimation: Each data point used for both training and testing
• Reduced Overfitting: Multiple validation rounds increase reliability
• Efficient Data Use: All data utilized for both training and validation

Mnemonic: "K-fold Keeps Keen Knowledge" (KKKK)

---

Question 3(a OR) [3 marks]

Give any three applications of multiple linear regression.

Answer:

Multiple Linear Regression Applications:

• Stock Market Analysis: Multiple economic indicators predict stock prices
• Academic Performance: Study hours, attendance, previous grades predict scores
• Marketing ROI: Various marketing channels impact sales revenue

Mnemonic: "Houses, Sales, Medicine Predict Multiple Variables" (HSMPV)

---

Question 3(b OR) [4 marks]

Find Standard Deviation for the following data: 4,15,20,28,35,45

Answer:

Standard Deviation Calculation:

Detailed Calculation:

• Deviations from mean: (-20.5)², (-9.5)², (-4.5)², (3.5)², (10.5)², (20.5)²
• Squared deviations: 420.25, 90.25, 20.25, 12.25, 110.25, 420.25
• Sum: 1073.5
• Variance: 1073.5/6 = 178.92
• Standard Deviation: √178.92 = 13.38

Mnemonic: "Deviation Measures Data Spread" (DMDS)

---

Question 3(c OR) [7 marks]

Explain Bagging, Boosting in detail.

Answer:

Ensemble Methods Comparison:

Bagging (Bootstrap Aggregating):

Boosting Process:

• Sequential Learning: Each model learns from previous model's mistakes
• Weight Adjustment: Increase weight of misclassified examples
• Final Prediction: Weighted combination of all models

Key Differences:

• Bagging: Independent models trained in parallel, reduces overfitting
• Boosting: Dependent models trained sequentially, improves accuracy

Mnemonic: "Bagging Builds Parallel, Boosting Builds Sequential" (BBPBS)

---

Question 4(a) [3 marks]

Define: Support, Confidence.

Answer:

Association Rule Metrics:

Example:

• Support(Bread) = 0.6 (60% transactions contain bread)
• Confidence(Bread→Butter) = 0.8 (80% of bread buyers also buy butter)

Applications:

• Market Basket Analysis: Finding product associations
• Recommendation Systems: Suggesting related items

Mnemonic: "Support Shows Frequency, Confidence Shows Connection" (SSFC)

---

Question 4(b) [4 marks]

Illustrate any two applications of logistic regression.

Answer:

Logistic Regression Applications:

Key Features:

• Binary Classification: Predicts probability between 0 and 1
• S-shaped Curve: Uses sigmoid function for probability estimation
• Linear Decision Boundary: Separates classes with linear boundary

Real-world Examples:

• Marketing: Customer purchase probability based on demographics
• Finance: Credit approval based on credit history and income

Mnemonic: "Logistic Limits Linear Logic" (LLLL)

---

Question 4(c) [7 marks]

Discuss the main purpose of Numpy and Pandas in machine learning.

Answer:

NumPy and Pandas in ML:

NumPy Functions:

Pandas Capabilities:

• Data Import/Export: Read CSV, Excel, JSON files
• Data Cleaning: Handle missing values, duplicates
• Data Transformation: Group, merge, pivot operations
• Statistical Analysis: Descriptive statistics, correlation

Integration with ML:

• Data Preprocessing: Clean and prepare data for algorithms
• Feature Engineering: Create new features from existing data
• Model Input: Convert data to format required by ML algorithms

Key Benefits:

• Performance: Optimized C/C++ backend for speed
• Memory Efficiency: Efficient data storage and manipulation
• Ecosystem Integration: Works seamlessly with scikit-learn, matplotlib

Mnemonic: "NumPy Numbers, Pandas Processes Data" (NNPD)

---

Question 4(a OR) [3 marks]

Give any three examples of Supervised Learning.

Answer:

Supervised Learning Examples:

• Medical Diagnosis: Patient symptoms → Disease classification
• Stock Price Prediction: Market indicators → Future price
• Speech Recognition: Audio signals → Text transcription

Mnemonic: "Emails, Houses, Images Learn Supervised" (EHILS)

---

Question 4(b OR) [4 marks]

Explain any two applications of the apriori algorithm.

Answer:

Apriori Algorithm Applications:

Market Basket Analysis:

• Example: "Customers who buy bread and milk also buy eggs"
• Business Impact: Product placement, promotional offers
• Implementation: Analyze transaction data to find frequent itemsets

Web Usage Mining:

• Example: "Users visiting page A often visit page B next"
• Website Optimization: Improve navigation, recommend content
• User Experience: Personalized website layouts

Algorithm Process:

• Generate Candidates: Create frequent itemsets
• Prune: Remove infrequent items
• Generate Rules: Create association rules with confidence

Mnemonic: "Apriori Analyzes Associations Automatically" (AAAA)

---

Question 4(c OR) [7 marks]

Explain the features and applications of Matplotlib.

Answer:

Matplotlib Features and Applications:

Key Features:

Applications in Machine Learning:

• Data Exploration: Visualize data distribution and patterns
• Model Performance: Plot accuracy, loss curves during training
• Result Presentation: Display predictions vs actual values
• Feature Analysis: Correlation matrices, feature importance plots

Advanced Capabilities:

• Animation: Create animated plots for time-series data
• Interactive Widgets: Add sliders, buttons for user interaction
• Integration: Works with Jupyter notebooks, web applications

Benefits:

• Flexibility: Highly customizable plotting options
• Community: Large user base with extensive documentation
• Compatibility: Integrates with NumPy, Pandas seamlessly

Mnemonic: "Matplotlib Makes Meaningful Visual Displays" (MMVD)

---

Question 5(a) [3 marks]

List out the major features of Numpy.

Answer:

NumPy Major Features:

• Universal Functions: Element-wise operations on arrays
• Memory Efficiency: Contiguous memory layout for speed
• C/C++ Integration: Interface with compiled languages

Mnemonic: "NumPy Numbers Need Neat Operations" (NNNNO)

---

Question 5(b) [4 marks]

How to load an iris dataset csv file in a Pandas Dataframe program? Explain with example.

Answer:

Loading Iris Dataset:

Python

Code Explanation:

• pd.read_csv(): Reads CSV file into DataFrame
• columns parameter: Assigns column names
• head(): Shows first 5 rows
• info(): Displays data types and memory usage

Mnemonic: "Pandas Reads CSV Files Easily" (PRCFE)

---

Question 5(c) [7 marks]

Compare and Contrast Supervised Learning and Unsupervised Learning.

Answer:

Comprehensive Comparison:

Detailed Comparison:

Supervised Learning Characteristics:

• Training Process: Learn from examples with known correct answers
• Performance Measurement: Direct comparison with actual outcomes
• Common Algorithms: Decision trees, SVM, neural networks
• Business Applications: Fraud detection, medical diagnosis, price prediction

Unsupervised Learning Characteristics:

• Exploration: Find unknown patterns without guidance
• Validation Challenges: No ground truth for direct comparison
• Common Algorithms: K-means, hierarchical clustering, PCA
• Business Applications: Customer segmentation, market research, anomaly detection

Key Contrasts:

• Feedback: Supervised has immediate feedback, unsupervised relies on domain expertise
• Data Requirements: Supervised needs expensive labeled data, unsupervised uses readily available unlabeled data
• Problem Types: Supervised solves prediction problems, unsupervised solves discovery problems

Mnemonic: "Supervised Seeks Specific Solutions, Unsupervised Uncovers Unknown" (SSSUU)

---

Question 5(a OR) [3 marks]

List out the applications of Pandas.

Answer:

Pandas Applications:

• Scientific Research: Experimental data analysis
• Web Analytics: Website traffic, user behavior analysis
• Healthcare: Patient records, clinical trial data

Mnemonic: "Pandas Processes Data Perfectly" (PPDP)

---

Question 5(b OR) [4 marks]

How to plot a vertical line and horizontal line in matplotlib? Explain with examples.

Answer:

Matplotlib Line Plotting:

Python

Key Functions:

• axvline(): Creates vertical line at specified x-coordinate
• axhline(): Creates horizontal line at specified y-coordinate
• Parameters: color, linestyle, linewidth, alpha

Mnemonic: "Matplotlib Makes Lines Easily" (MMLE)

---

Question 5(c OR) [7 marks]

Describe the concept of clustering using appropriate real-world examples.

Answer:

Clustering Concept and Applications:

Clustering Process:

Detailed Examples:

1. Customer Segmentation:

• Data: Purchase history, demographics, website behavior
• Clusters: High-value customers, price-sensitive buyers, occasional shoppers
• Business Value: Customized marketing, product recommendations, retention strategies

2. Social Media Analysis:

• Data: User interactions, post topics, engagement patterns
• Clusters: Influencers, casual users, brand advocates
• Applications: Viral marketing, content strategy, community management

3. Market Research:

• Data: Survey responses, product preferences, demographics
• Clusters: Market segments with similar needs
• Insights: Product development, pricing strategy, market positioning

Clustering Algorithms:

• K-Means: Partitions data into k clusters
• Hierarchical: Creates tree-like cluster structure
• DBSCAN: Finds clusters of varying density

Validation Methods:

• Silhouette Score: Measures cluster quality
• Elbow Method: Determines optimal number of clusters
• Domain Expertise: Business knowledge validation

Benefits:

• Pattern Discovery: Reveals hidden data structures
• Decision Support: Provides insights for business decisions
• Automation: Reduces manual data analysis effort

Mnemonic: "Clustering Creates Clear Categories" (CCCC)