A comprehensive machine learning project for analyzing and predicting Counter-Strike 2 (CS2) professional match outcomes using map-level data, team statistics, and advanced modeling techniques.
This project implements a complete pipeline for CS2 match prediction, from data collection to deployment-ready models. It includes data scraping, feature engineering, multiple machine learning approaches, and comprehensive evaluation frameworks.
- Data Scraping: Automated collection from BO3.gg API
- Feature Engineering: Chronological winrates, recent form, and contextual features
- Dataset Extension: Automated updates with new tournament data
- Classification: Binary win/loss prediction with multiple algorithms
- Regression: Probabilistic win prediction with calibration
- Model Comparison: Comprehensive evaluation across multiple metrics
- Calibration Analysis: Probability reliability assessment
- Final Report: Complete scientific analysis and findings
🎯 Comprehensive ML Pipeline
- Multiple algorithms: Logistic Regression, Random Forest, XGBoost, LightGBM, MLP
- Advanced preprocessing with temporal awareness to prevent data leakage
- Automated hyperparameter tuning and model selection
📊 Rich Feature Engineering
- Chronological team performance metrics
- Map-specific winrates and preferences
- Head-to-head statistics and recent form
- Pick/ban context and strategic indicators
🔄 Automated Data Management
- Real-time tournament detection and dataset updates
- Duplicate prevention and data quality controls
- Backup systems and error handling
📈 Advanced Evaluation
- Multiple evaluation metrics (ROC-AUC, Brier Score, Log-Loss)
- Calibration analysis and reliability assessment
- Per-map model specialization and routing
pip install -r Dataset/requirements.txtcd Dataset
python cs2_match_data_scraper.py # Collect raw match data
python create_ml_dataset.py # Create ML-ready dataset
python extend_dataset.py # Update with new tournaments (optional)cd Evaluation/Classification
# Open and run cs2_map_prediction.ipynb for classification models
cd ../Regression
# Open and run cs2_map_probability_regression.ipynb for regression models- Source: Professional CS2 matches from BO3.gg API
- Timeframe: Last 6 months of tournament data
- Scale: 2,400+ map records from 1,000+ matches
- Teams: Top 50 teams by earnings
- Features: 20+ engineered features including temporal dynamics
| Model | ROC-AUC | Brier Score | Log-Loss | Accuracy |
|---|---|---|---|---|
| LightGBM | 0.72 | 0.231 | 0.659 | 67.2% |
| Random Forest | 0.71 | 0.233 | 0.665 | 66.8% |
| XGBoost | 0.70 | 0.235 | 0.672 | 66.1% |
| MLP | 0.69 | 0.238 | 0.675 | 65.7% |
├── Dataset/ # Data collection and preprocessing
│ ├── cs2_match_data_scraper.py # Main data scraper
│ ├── create_ml_dataset.py # ML dataset transformation
│ ├── extend_dataset.py # Automated dataset updates
│ ├── dataset_profiler.py # Data quality analysis
│ └── cs2api/ # Custom API library
├── Evaluation/ # Machine learning models
│ ├── Classification/ # Classification models and analysis
│ ├── Regression/ # Regression models and calibration
└── └── compilated/ # Results aggregation and reporting
- Temporal Awareness: All features computed chronologically to prevent data leakage
- Advanced Calibration: Probability reliability optimization for betting applications
- Per-Map Specialization: Custom models for different map types and contexts
- Robust Evaluation: Multiple train/validation/test strategies with proper grouping
- Production Ready: Automated pipelines with comprehensive error handling
This project was developed as part of a machine learning course focusing on real-world applications of predictive modeling in esports analytics.
MIT License - See LICENSE file for details
- BO3.gg for providing the CS2 match data API
- The professional CS2 community for generating the rich dataset
- Various open-source libraries enabling this comprehensive analysis