Project 3: Titanic Survival Prediction

Predict which passengers survived the Titanic disaster using passenger information.

Difficulty

Beginner

Time Estimate

2-3 days

Skills You'll Practice

• Classification
• Feature Engineering
• Data Cleaning
• Handling Missing Values
• Categorical Encoding

Learning Objectives

By completing this project, you will learn to:

• Handle missing data effectively
• Encode categorical variables
• Engineer features from raw data
• Apply multiple classification algorithms
• Evaluate classification models
• Handle imbalanced datasets

Dataset

Kaggle Titanic Competition

• Titanic - Machine Learning from Disaster
• Download train.csv and test.csv
• Classic beginner-friendly dataset

Project Steps

Step 1: Load and Explore Data

• Load train and test datasets
• Check data shape and basic statistics
• Identify missing values
• Explore feature distributions
• Analyze survival rates by different features

Step 2: Data Preprocessing

• Handle missing values (Age, Cabin, Embarked)
• Create new features (Family Size, Title, etc.)
• Encode categorical variables (Sex, Embarked)
• Remove irrelevant features

Step 3: Feature Engineering

• Extract title from Name
• Create family size from SibSp and Parch
• Create age groups
• Create fare groups
• Handle outliers

Step 4: Model Training

• Split data into train/validation sets
• Train multiple models:
  • Logistic Regression
  • Random Forest
  • Decision Tree
  • K-Nearest Neighbors
• Compare model performance

Step 5: Model Evaluation

• Calculate accuracy, precision, recall, F1-score
• Create confusion matrix
• Plot ROC curve
• Analyze feature importance

Step 6: Model Improvement

• Tune hyperparameters
• Try ensemble methods
• Feature selection
• Final model selection and submission

Expected Deliverables

1. Jupyter Notebook with complete analysis:

   • EDA with visualizations
   • Data preprocessing steps
   • Feature engineering
   • Model training and evaluation
   • Results and conclusions

2. Submission File:

   • CSV file with predictions for test set
   • Format: PassengerId, Survived

Evaluation Metrics

• Accuracy: Overall correctness
• Precision: Of predicted survivors, how many actually survived
• Recall: Of actual survivors, how many were found
• F1-Score: Harmonic mean of precision and recall

Key Features to Explore

• Pclass: Passenger class (1, 2, 3)
• Sex: Gender (male, female)
• Age: Age of passenger
• SibSp: Number of siblings/spouses aboard
• Parch: Number of parents/children aboard
• Fare: Ticket fare
• Embarked: Port of embarkation
• Cabin: Cabin number (many missing)

Feature Engineering Ideas

• Extract title from Name (Mr, Mrs, Miss, etc.)
• Create family size = SibSp + Parch + 1
• Create is_alone flag
• Create age groups (child, adult, senior)
• Create fare groups
• Extract deck from Cabin (if available)

Tips

• Start with simple models (Logistic Regression)
• Visualize survival rates by different features
• Handle missing Age carefully (use median or predict)
• Sex is a very important feature
• Try creating new features from existing ones
• Submit to Kaggle to see your score!

Resources

• Kaggle Titanic Competition
• Titanic Dataset on Kaggle
• Scikit-learn Classification

Next Steps

After completing this project:

• Submit predictions to Kaggle
• Try advanced feature engineering
• Experiment with ensemble methods
• Move to Project 4: Spam Detection