StockPrediction

Advanced stock prediction and trading system using machine learning with walk-forward validation and multi-stock training.

🚀 Features

• 80+ Normalized Features: Return-based indicators that generalize across equities.
• Walk-Forward Validation: Simulates real trading conditions to prevent overfitting.
• Multi-Stock Training: Leverages patterns across multiple tickers (e.g., SPY, QQQ, AAPL).
• Financial Metrics: Optimizes for Sharpe ratio, win rate, and profit factor.
• Model Ensemble: Combines Random Forest, XGBoost, LightGBM, and GBM with a Logistic Regression meta-model.
• Interactive Dashboard: Real-time backtesting, metrics visualization, and next-day forecasts.

📁 Project Structure

• features_engineering.py: Technical indicators and return-based feature generation.
• modeling.py: Walk-forward validator and ensemble model architecture.
• training.py: Script for training on single or multiple stocks.
• streamlit_dashboard.py: Streamlit-based user interface.
• models/: Directory for saved .pkl model files and metrics.

🎓 Quick Start

1. Install Dependencies

brew install ta-lib # Required for TA-Lib
pip install -r requirements.txt

2. Set Up Environment

Create a .env file with your API keys:

ALPHA_VANTAGE_API_KEY=your_key_here

3. Train and Launch

# Train on major ETFs
python training.py --mode multi --symbols SPY QQQ IWM DIA

# Launch the dashboard
streamlit run streamlit_dashboard.py

📈 Technical Pipeline

1. Preprocessing: Data download via yfinance and sentiment fetching via Alpha Vantage.
2. Feature Engineering: 80+ normalized features including volatility, momentum, and volume ratios.
3. Training: Expanding-window walk-forward validation with a 5-day gap to eliminate look-ahead bias.
4. Ensemble: Weighted stacking of multiple gradient-boosted models.
5. Backtesting: Transaction cost modeling and monthly performance analysis.

---

⚠️ Disclaimer

Educational and research purposes only. Trading stocks involves risk. Past performance does not guarantee future results.