🛒 E-commerce Session Value Prediction - Datathon 2025

📋 Project Overview

This project focuses on predicting the monetary value of user sessions in an e-commerce platform using machine learning techniques. The goal is to analyze user behavior patterns and build a robust predictive model that can estimate session values based on user interactions.

🎯 Objective

Predict the session_value for e-commerce user sessions using behavioral event data including views, cart additions, and purchases.

🏆 Competition Context

• Event: Datathon 2025
• Task Type: Regression (Session Value Prediction)
• Evaluation Metric: MSE (Mean Squared Error) and RMSE
• Data Format: Time-series user event data

📊 Dataset Description

Training Data (train (1).csv)

• User Events: Individual user interactions (views, cart additions, purchases)
• Session Information: User session identifiers and timestamps
• Product Details: Product and category identifiers
• Target Variable: session_value - monetary value of each session

Test Data (test (1).csv)

• Same structure as training data
• Missing session_value (target for prediction)
• Used for final model evaluation and submission

Key Features

• user_id: Unique user identifier
• user_session: Unique session identifier
• event_type: Type of user action (VIEW, ADD_CART, BUY)
• product_id: Unique product identifier
• category_id: Product category identifier
• event_time: Timestamp of the event
• session_value: Target variable (monetary value)

🔧 Technical Approach

1. Exploratory Data Analysis (EDA)

• Data Quality Assessment: Missing value analysis and data integrity checks
• Target Distribution Analysis: Understanding session value patterns and skewness
• Feature Distribution: Analyzing event types, product categories, and user behaviors
• Temporal Patterns: Event sequence analysis and time-based insights
• Cross-validation: Train/test distribution comparison for model generalization

2. Feature Engineering

• Temporal Features: Hour, day of week, month, weekend indicators
• Cyclical Encoding: Sine/cosine transformations for time periodicity
• Session Aggregations: Event counts, unique products/categories per session
• Behavioral Metrics: Event type ratios and conversion indicators
• User Patterns: Historical user behavior and engagement metrics

3. Machine Learning Pipeline

• Model Selection: CatBoost Regressor (optimized for categorical features)
• Data Splitting: Time-based validation (80/20 split)
• Target Transformation: Log1p transformation to handle skewed distribution
• Hyperparameter Optimization: Optuna-based automated tuning
• Validation Strategy: Early stopping and cross-validation

4. Model Optimization

• Automated Hyperparameter Tuning: 20 trials with Optuna optimization
• Parameters Optimized: iterations, learning_rate, depth, regularization
• Categorical Handling: Native CatBoost categorical feature processing
• Overfitting Prevention: Early stopping and validation monitoring

📁 Project Structure

datahon25/
├── README.md                          # Project documentation
├── notebook9db22db4c8.ipynb          # Main analysis notebook
├── train (1).csv                     # Training dataset
├── test (1).csv                      # Test dataset
├── catboost_submission.csv           # Final predictions
└── 520.csv                           # the csv that get 523 score on private lb

🚀 Getting Started

Prerequisites

# Required Python packages
pip install numpy pandas matplotlib seaborn missingno scikit-learn catboost optuna

Running the Analysis

1. Clone the repository

   git clone https://github.com/YsK-dev/datahon25.git
   cd datathon25

2. Open the Jupyter Notebook

   jupyter notebook notebook9db22db4c8.ipynb

3. Run the complete pipeline

   • Execute cells sequentially for full analysis
   • EDA section provides comprehensive data insights
   • ML pipeline generates final predictions

Key Notebook Sections

• 📦 Package Installation: Required dependencies
• 🔍 Data Loading & Exploration: Initial data analysis
• 🎯 Target Variable Analysis: Session value distribution
• 🔄 Event Type Analysis: User behavior patterns
• 🛍️ Product & Category Analysis: Market insights
• 👥 User Behavior Analysis: Engagement patterns
• ⏰ Temporal Analysis: Time-based patterns
• 🔧 Feature Engineering: Advanced feature creation
• 🤖 Machine Learning Pipeline: Model training and optimization
• 📊 Results & Submission: Final predictions and evaluation

📈 Results & Performance

Model Performance

• Algorithm: CatBoost Regressor with Optuna optimization
• Validation Strategy: Time-based split (chronological)
• Target Transformation: Log1p (handles skewed distribution)
• Feature Count: 15+ engineered features
• Categorical Features: Native CatBoost handling

Key Achievements

• ✅ Comprehensive EDA with actionable insights
• ✅ Robust feature engineering pipeline
• ✅ Automated hyperparameter optimization
• ✅ Time-aware validation strategy
• ✅ Production-ready prediction pipeline

Output Files

• catboost_submission.csv: Final predictions for test data
• Session-level predictions with quality validation
• Duplicate detection and data integrity checks

🔍 Key Insights

Data Patterns

• Event Distribution: Majority of events are product views
• Session Behavior: Longer sessions tend to have higher values
• Temporal Patterns: Clear time-of-day and day-of-week effects
• User Segments: Distinct high-value and casual user groups

Feature Importance

• Session event counts and engagement metrics
• Event type ratios (especially purchase/cart ratios)
• Temporal features (hour, day patterns)
• Product category preferences
• User historical behavior patterns

🛠️ Technical Details

Model Architecture

• Base Model: CatBoost Regressor
• Categorical Features: 5 native categorical features
• Numerical Features: 10+ engineered numerical features
• Target Processing: Log-normal distribution handling

Optimization Strategy

• Hyperparameter Search: Bayesian optimization with Optuna
• Search Space: 7 key parameters optimized
• Evaluation Metric: RMSE on validation set
• Early Stopping: Prevents overfitting

Data Processing

• Missing Value Handling: Comprehensive imputation strategy
• Categorical Encoding: Native CatBoost processing
• Feature Scaling: Log transformation for target variable
• Temporal Features: Cyclical encoding for time variables

📚 Dependencies

Core Libraries

• pandas (1.5+): Data manipulation and analysis
• numpy (1.21+): Numerical computations
• scikit-learn (1.1+): Machine learning utilities
• catboost (1.0+): Gradient boosting framework

Visualization

• matplotlib (3.5+): Basic plotting
• seaborn (0.11+): Statistical visualizations
• missingno: Missing data visualization

Optimization

• optuna (3.0+): Hyperparameter optimization

🤝 Contributing

1. Fork the repository
2. Create a feature branch (git checkout -b feature/improvement)
3. Commit your changes (git commit -am 'Add new feature')
4. Push to the branch (git push origin feature/improvement)
5. Create a Pull Request

📄 License

This project is part of Datathon 2025 competition. Please respect competition rules and academic integrity guidelines.

👨‍💻 Author

YsK-dev

• GitHub: @YsK-dev
• Project: Datathon 2025 - E-commerce Session Value Prediction

🔗 Additional Resources

• CatBoost Documentation
• Optuna Optimization Guide
• Scikit-learn User Guide

---

Last Updated: September 2025 Competition: Datathon 2025