This project aims to predict stock market buy/sell signals using a wide array of technical indicators and machine learning models, targeting traders and quantitative analysts. It offers a comprehensive framework for backtesting and real-time prediction, potentially improving trading strategies.

How It Works

The system processes historical stock data (OHLCV) to generate a large set of technical indicators (over 600 claimed). It then trains and evaluates up to 36 different machine learning models, including TensorFlow (LSTM, GRU, Dense) and XGBoost, to predict categorical buy/sell/hold signals. Feature selection is performed to identify the most relevant indicators for each stock, and a real-time evaluation system provides alerts via Telegram.

Quick Start & Requirements

• Install dependencies: pip install -r requirements.txt
• Data collection: Requires AlphaVantage or Alpaca API keys for historical data.
• Python version: Recommended Python 3.8; caution advised with newer versions due to potential pandas compatibility issues.
• Hardware: Estimated setup time is significant, with model training taking ~15 minutes per stock.

Highlighted Details

• Tests up to 36 models, including TensorFlow and XGBoost.
• Utilizes over 600 technical stock indicators.
• Provides categorical buy/sell/do nothing predictions.
• Features a real-time evaluation and alerting system (Telegram, Mail).
• Includes detailed tutorials and extensive "possible improvements" sections.

Maintenance & Community

The project is actively seeking collaborators. A Telegram group is available for community interaction and coordination.

Licensing & Compatibility

The project is offered "free" with the condition that major improvements are shared with the author. Commercialization without authorization is prohibited.

Limitations & Caveats

The project is described as "long and dense," requiring careful understanding of the tutorial before installation. The author notes that stock market prediction is inherently difficult due to external factors not captured by technical indicators. Real-time data collection for volume is noted as challenging, and the project is not recommended for intervals less than 1 hour. The README also contains a disclaimer stating "USE THE SOFTWARE AT YOUR OWN RISK."