📈 CAP4630_Final_Project – Stock Forecasting Ensemble

This project is a Streamlit-based web application for 30-day stock price forecasting using an ensemble of deep learning models and statistical methods.

Each team member contributed a model architecture, enabling the app to compare performance across various forecasting techniques. Users can input a stock ticker, and the app will train or load models, visualize predictions, and allow CSV download of forecasted prices.

---

🧠 Team Model Contributions

Member	Model Architecture
Juan	LSTM Seq2Seq
Omar	LSTM Bidirectional Seq2Seq
Jayvee	LSTM Bidirectional Regressor
Tito	GRU Seq2Seq
Luis	ARIMA (Statistical) & Transformer Regressor

---

⚙️ Setup Instructions

1️⃣ Clone the repository

git clone https://github.com/your-username/CAP4630_Final_Project.git
cd CAP4630_Final_Project
git checkout dev

2️⃣ Create a virtual environment using pip

On Windows:

python -m venv project_env
project_env\Scripts\activate

On macOS/Linux:

python3 -m venv project_env
source project_env/bin/activate

---

3️⃣ Install the required Python packages

Make sure you're inside the root folder where requirements.txt is located:

pip install -r requirements.txt

---

4️⃣ Run the Streamlit app

streamlit run app.py

Then visit http://localhost:8501 in your browser.

---

🖼️ App Features

• ✅ Input any stock ticker (e.g. TSLA, AAPL, GOOG)
• 📊 Train models or load pre-trained ones
• 🧠 Uses LSTM, GRU, and Transformer architectures
• 📈 Visualize forecasts in a multi-line Plotly chart
• 💾 Download predictions as CSV
• 🚀 Efficient training with session caching and model persistence

---

📁 Project Structure

CAP4630_Final_Project/
│
├── app.py                    # Streamlit UI
├── ensemble_forecasts.py     # Model training, forecasting, and utilities
├── requirements.txt          # Python dependencies
├── models/                   # Saved model files organized by ticker
├── plots/                    # Saved visualizations (optional)
└── README.md                 # This file

---

🧩 Dependencies

Make sure your environment includes:

• streamlit
• pandas
• numpy
• plotly
• tensorflow
• scikit-learn
• statsmodels
• yfinance
• stqdm (optional, for progress bars)

You don’t need to install these manually — they are all in requirements.txt.

---

🧠 Technologies Used

• TensorFlow / Keras – for building LSTM, GRU, and Transformer models
• Statsmodels – for classical ARIMA forecasting
• Plotly – for interactive line charts
• Streamlit – for building the user interface
• scikit-learn – for metrics and preprocessing
• yfinance – to fetch historical stock prices

---

🎓 About the Course

This project was created for CAP4630: Introduction to Artificial Intelligence at Florida International University.

---

👥 Authors

• Juan – LSTM Seq2Seq
• Omar – LSTM Bidirectional Seq2Seq
• Jayvee – LSTM Bidirectional Regressor
• Tito – GRU Seq2Seq
• Luis – ARIMA & Transformer Forecasting

---

📌 Notes

• Models are saved inside /models/<TICKER>/ as .keras files.
• The app will reuse existing models if available, speeding up repeat forecasts.
• Forecasts are always for the next 30 days from the last date in the dataset.

---

Enjoy exploring your forecasts! 📊📈💡