A full-stack application that predicts TSMC's next-day closing stock price using deep learning models and visualizes forecasts via an interactive React UI.
- Languages: Python, JavaScript, CSS
- Frameworks/Libraries: Flask, React, Vite, TensorFlow/Keras, Pandas, Plotly
- Tools/Platforms: Docker, Docker Compose, Google Cloud Run, Firebase Hosting, GitHub Actions
- Multi-Model Forecasting: Evaluates and compares LSTM, GRU, Conv1D, and FFN architectures for time series prediction.
- Interactive Visualization: A modern React frontend using Plotly to visualize validation and test predictions against true stock prices.
- Full-Stack Deployment: Containerized backend on Google Cloud Run and static frontend hosting on Firebase, orchestrated via Docker for local development.
- Automated CI/CD: GitHub Actions workflows ensure code quality and automated testing upon every push.
Live Demo: https://time-series-backend.web.app/
The FFN and GRU models demonstrated the most reliable generalization to unseen data.
- FFN: Test MAE: $4.24 (2.19%) - Best generalization
- GRU: Test MAE: $4.27 (2.21%) - Balanced fit
- LSTM: Test MAE: $6.56 (3.40%) - Showed overfitting
# Clone the repository
git clone https://github.com/hungkaihsin/tsmc_stock_forecasting.git
cd tsmc_stock_forecasting
# Build and Start with Docker Compose
docker-compose build
docker-compose up
# Access the application
# Open your browser to http://localhost:8080Created by Daniel - [email protected] | LinkedIn | Portfolio