CS:GO AI Match Predictor (Time-Travel Logic)

An advanced Machine Learning project that predicts the winner of professional CS:GO matches using historical data, ELO ratings, and momentum analysis. Built with XGBoost and engineered to prevent data leakage.

Key Features

• XGBoost Engine: Optimized gradient boosting model with robust hyperparameters found via exhaustive search.
• Time-Travel Simulation: The system fetches team stats (ELO, Rank, Form) exactly as they were on the specific date of the last meeting, ensuring 100% realistic historical comparison.
• Symmetric Training: Implements data mirroring to eliminate positional bias (Team A vs B is treated same as Team B vs A).
• Pistol Round Impact: Analyzes historical pistol round win rates (Round 1 & 16) as a key momentum indicator.
• Risk Analysis: Provides a confidence score and risk warning (e.g., "High Risk / Coin Flip") for close matchups.
• Leakage-Free: Rigorous cleaning of "future" features to ensure model integrity.

Performance

• Accuracy: ~78% (on 2019-2020 Test Data)
• Methodology: XGBoost Classification with TimeSeries Split and custom feature engineering (ELO, KAST, WR).

Installation & Setup

1. Clone the repository:

   git clone http://github.com/Islingr5/CSGO-Prediction-Project.git
   cd CSGO-Prediction-Project

   (Note: This repository tracks large binary files using Git LFS. Ensure you have Git LFS installed to fetch the model and datasets.)

2. Install dependencies:

   pip install -r requirements.txt

   (Note: I suggest you to use Python's virtual environment "venv" for pip usage in linux systems.)

3. Authentication Check: Ensure you have correctly configured your Git Personal Access Token (PAT) for clone operations.

💻 Usage & Execution

The project requires an initial training run to process all historical data and save the final model (csgo_v11_model.pkl).

1. Initial Training & Model Generation (Run Once)

This script loads all raw data, generates features, trains the model on the full historical dataset (2015-2018), and validates its performance against the test set (2019-2020).

python run.py

💻 Usage & Execution

2. Interactive Prediction (Demo Mode)

After the model file (csgo_v11_model.pkl) is generated, you can use the built-in interactive shell to perform instant simulations.

• Execution: Loads the saved model instantly for predictions.

  python demo.py

Interactive Commands:

• list: Shows all available team names (paginated).
• maplist: Shows all available maps (e.g., Mirage, Dust2, Inferno).
• Input Logic: The system will prompt you for two teams and a map. It will then perform a Time-Travel Simulation by finding the last historical meeting between those teams and predicting the winner based on the stats of that specific day.