Classification Exercise

What you'll learn

In this hands-on activity, you will:

• Master three fundamental classification algorithms (KNN, Naive Bayes, Decision Trees)
• Apply classification techniques to real-world datasets
• Compare and evaluate different classification models
• Handle imbalanced datasets in fraud detection
• Build predictive models for healthcare applications

:computer: Access the exercise template here: Classification Exercise Template

Exercise Overview

Part One: Small Sample Dataset :dart: (20-25 minutes)

In this section, you will:

1. Implement K-Nearest Neighbors (KNN) classification
2. Apply Naive Bayes algorithm for classification
3. Build Decision Tree classifiers
4. Compare performance across different algorithms
5. Understand the strengths and weaknesses of each method

:bulb: Start with KNN first! It's the most intuitive algorithm. Try k=3 and k=5 to see how the number of neighbors affects your results.

Part 2: Iris Flower Dataset :emoji: (15-20 minutes)

You will learn to:

1. Work with the classic Iris dataset
2. Apply classification techniques to real-world botanical data
3. Perform feature analysis and visualization
4. Implement multi-class classification
5. Evaluate model performance on standard datasets tip Visualization tip: Create scatter plots of sepal length vs. sepal width to see how well the different species separate. This will help you understand why certain algorithms perform better.

Part 3: Advanced Challenge 1 :rocket: (30-35 minutes)

This section covers two real-world applications:

Challenge 1 - Heart Disease Prediction :emoji:

1. Load and explore medical data
2. Implement binary classification for disease prediction
3. Analyze feature importance in medical diagnosis
4. Apply appropriate preprocessing for healthcare data

:bulb: Important: Medical data often has missing values. Check for NaN values and handle them appropriately before training your models.

Challenge 2 - Payment Fraud Detection :emoji:

1. Load and analyze financial transaction data
2. Handle imbalanced datasets (fraud cases are rare!)
3. Implement cost-sensitive classification
4. Evaluate using appropriate metrics for imbalanced data tip Imbalanced data warning: Don't rely only on accuracy! With fraud detection, precision and recall are much more important. A model that predicts "no fraud" for everything might have 99% accuracy but be completely useless.

Part 4: Advanced Challenge 2 :trophy: (20-25 minutes)

You will master:

1. Complex classification scenarios with multiple classes
2. Advanced model optimization techniques
3. Ensemble methods and model combination
4. Production-ready classification systems

:bulb: Tip Pro tip: Try combining different classifiers using voting. Sometimes an ensemble of simple models outperforms a single complex model!

Exercise Submission :emoji:

:warning: Warning IMPORTANT: Submit your completed notebook here:

:link: Exercise Submission Form

Deadline: Submit before the next class session

Submission Guidelines

:emoji:️ Before submitting, ensure you have:

1. :white_check_mark: Completed all questions in each part
2. :white_check_mark: Verified your code runs without errors (Run All cells!)
3. :white_check_mark: Included proper model evaluation and comparison
4. :white_check_mark: Provided clear interpretations of classification results
5. :white_check_mark: Saved your notebook with a descriptive filename

Troubleshooting Common Issues

Import errors?

• Make sure you run the import cell at the beginning of the notebook
• If using Google Colab, restart runtime if packages don't load

Model not training?

• Check if your data is properly loaded
• Verify you've split your data into training and testing sets
• Ensure features and labels are correctly separated

Submission form not working?

• Try a different browser or incognito mode
• Make sure your file is saved as .ipynb format
• Contact your instructor if issues persist

:information_source: Info :bulb: Need help? Post questions in the class discussion forum or attend office hours. Classification can be tricky, but we're here to help!