This project focuses on stock price prediction using a transformer-based architecture. The model is trained on stock data retrieved via the yfinance library, applying various technical indicators to enhance predictive accuracy.
The model is trained on data from the following stocks:
- Apple (AAPL)
- Microsoft (MSFT)
- Amazon (AMZN)
- Meta (META)
- Google (GOOGL)
The dataset includes stock data from the last 30 days, sampled at 5-minute intervals. Additionally, several technical indicators, such as RSI (Relative Strength Index), Bollinger Bands, and ROC (Rate of Change), were calculated and added to the dataset.
The dataset was pre-processed and stored in a NumPy array for training.
- TensorFlow LayerNormalization Issue: In TensorFlow 2.16, I encountered an issue with the
LayerNormalizationlayer. This was resolved by uninstalling TensorFlow and upgrading to version 2.17. - Hardware Limitations: Due to limited computing resources, I was unable to train the model for more than 10 epochs.
The model achieved a directional accuracy of 70% on both the training and validation sets. The overall training accuracy stands at 65%.
In the future, I plan to:
- Transition from TensorFlow to PyTorch to leverage its flexibility.
- Add more layers to the transformer architecture, making it more complex to better capture the intricacies of stock market data.
- Author: Mrunal Ashwinbhai Mania
- University: Arizona State University
- Email: [email protected]