zeroday01/ipl_prediction_engine

IPL Prediction Engine (Hybrid XGBoost + LSTM)

🌐 Live Demo

Hugging Face Space: zeroday01/ipl_prediction_engine

A high-performance machine learning engine designed to predict IPL match outcomes and ball-by-ball win probabilities using a hybrid ensemble architecture.

🚀 Key Features

• Hybrid Architecture: Combines XGBoost (for static match context) and LSTM (for temporal momentum).
• Enriched Context: Real-time integration of historical Open-Meteo Weather Data (dew, temperature, humidity).
• Dynamic Form Tracking: Rolling ELO ratings for teams and players calculated from 2008 to 2024.
• Memory Efficient: Optimized training pipeline using IterableDataset to handle 1000+ match files on low-RAM hardware.
• Real-Time Ready: Built-in support for Redis-based live match streaming.

🛠 Project Structure

• backend/ml_engine/hybrid_model.py: Core model architecture and normalization.
• backend/ml_engine/train_efficient.py: Memory-efficient training pipeline.
• backend/data_pipeline/fetch_weather_data.py: Historical weather data harvester.
• backend/data_pipeline/feature_engineer_elo.py: ELO and form calculation engine.

📦 Installation

1. Clone the repository.
2. Create a virtual environment:

   python -m venv .venv
   source .venv/bin/activate
   

3. Install dependencies:

   pip install -r requirements.txt
   

📈 Usage

1. Enrich the Data

First, generate the weather and ELO datasets:

python backend/data_pipeline/fetch_weather_data.py
python backend/data_pipeline/feature_engineer_elo.py

2. Train the Model

python backend/ml_engine/train_efficient.py

3. Run Inference Test

python backend/ml_engine/test_inference.py

⚖️ License

MIT