stocks_ML is a Python-based project that uses TensorFlow and Long Short-Term Memory (LSTM) neural networks to analyze stock time series data. The model is designed to predict which stocks are likely to perform well on the next day based on the previous day's data. This project does not provide financial advice or recommendations for buying or selling stocks. It is intended for educational and research purposes only.
- Utilizes LSTM networks for time series forecasting.
- Input: Historical stock data (previous day's data points such as open, close, high, low, volume).
- Output: Probability scores indicating which stocks are likely to perform well on the next day.
- Performance evaluation using standard metrics like accuracy and ROC-AUC curve.
- Python 3.x
- TensorFlow 2.x
- Pandas, NumPy for data manipulation.
- Matplotlib for data visualization and result plotting.
- yfinance for downloading stock data.
The model achieves approximately 80% accuracy in identifying stocks that perform exceptionally well (defined as a >4% increase in stock price in each of the first four hours of the trading day, compared to the same hour the previous day). This performance is based on test data after training on historical data using the LSTM model.
Below is the ROC curve of the model (binary_model_1695stocks_31days_4percent_2024_07_05), which is the most recent and best-performing model available in the repository.
This model demonstrates strong predictive power for identifying stocks likely to perform well, with high sensitivity and specificity as shown in the ROC curve.
Various trained models are saved as .keras files in the models/ directory.
- The file
ROC_binary_model_1695stocks_31days_4percent_2024_07_05.kerasis the best-performing model so far, with the highest accuracy and ROC-AUC score on the test data. - You can load these models for predictions or further fine-tuning.
To get started, clone the repository and install the required dependencies:
git clone https://github.com/noahhood/stocks_ML.git
cd stocks_ML
pip install -r requirements.txt