Stock Market Prediction with Ensemble Deep Learning

A sophisticated stock market prediction system that combines CNN and LSTM models to predict stock price movements. The model has been tested on various stocks and shows promising results in backtesting.

Features

• Ensemble Model Architecture: Combines CNN and Bidirectional LSTM for robust predictions
• Real-time Predictions: Uses Alpha Vantage API for live stock data
• Smart Caching: Implements efficient data caching to handle API rate limits
• Backtesting Engine: Complete backtesting system with visualization
• Adaptive Trading Strategy: Dynamic thresholds based on model confidence

Example Results

The above plot shows a successful trade where the model:

1. Bought at ₹2,710.35 when it predicted a relatively small drop (-4.382%)
2. Sold at ₹2,896.15 when it hit the take-profit target (+6.86%)
3. Generated a total return of 6.50% in a short period

Quick Start

1. Clone the repository

git clone https://github.com/yourusername/stock-market-prediction.git
cd stock-market-prediction

2. Install dependencies

pip install -r requirements.txt

3. Set up your environment

• Copy .env.sample to .env
• Get your API key from Alpha Vantage
• Add your API key to .env

4. Train the model

python train.py

This will:

• Download historical data for your chosen stock
• Train the CNN and LSTM models
• Create an ensemble model
• Save the trained model in models/ensemble_model.pth

Note: Each time you run training, it will create a fresh model optimized for your chosen stock and recent market conditions. This ensures your predictions are based on up-to-date patterns.

5. Run backtests

python backtest.py

6. Make predictions

python main.py --symbol RELIANCE.BSE

Model Architecture

The system uses an ensemble of two models:

1. CNN Model

• 1D Convolutions for pattern recognition
• LeakyReLU activation
• Batch normalization
• Dropout for regularization

2. LSTM Model

• Bidirectional LSTM for temporal patterns
• Sequence-to-one architecture
• Batch normalization
• Dropout for regularization

3. Ensemble Combination

• Weighted average of CNN and LSTM predictions
• Learnable weights with softmax normalization
• Small positive bias to counter bearish tendencies

Training Tips

1. Choose Your Stock: The model performs best when trained on the specific stock you want to predict
2. Data Range: Use at least 6-12 months of historical data for training
3. Validation: Always check the backtesting results before using predictions
4. Retraining: Consider retraining periodically to adapt to changing market conditions

Trading Strategy

The backtester implements a flexible trading strategy:

• Buy when predicted drop is less than threshold (-4.5%)
• Sell when predicted drop exceeds threshold (-6.5%)
• Stop loss at -1%
• Take profit at +1.5%

Data Handling

• Uses Alpha Vantage API for reliable stock data
• Implements smart caching to handle rate limits
• Normalizes features using percentage changes
• Handles missing data and outliers

Contributing

Feel free to:

• Fork the repository
• Create a feature branch
• Submit pull requests

License

MIT License - feel free to use this for your own projects!

Disclaimer

This is an experimental project. The predictions should not be used as financial advice. Always do your own research before making investment decisions.