StrataSight

StrataSight is a stock price forecasting application with full CRUD functionality that uses machine learning models to predict future stock prices. Built with Streamlit for an intuitive interface, it leverages yfinance for real-time market data, TensorFlow/Keras LSTM and Facebook's Prophet for forecasting, and Plotly for interactive visualizations.

Features

Prediction Generation

• Simple One-Click Predictions: Select a stock, choose forecast horizon, pick models, and generate predictions instantly
• Dual Model Approach:
  • LSTM (Long Short-Term Memory): Deep learning neural network for capturing complex patterns and short-term fluctuations
  • Prophet: Facebook's robust time series forecasting tool for stable trends and seasonal effects
• Interactive Visualizations: Plotly charts showing historical data and model predictions side-by-side
• Smart Model Caching: Models train once per stock and are cached for instant subsequent predictions
• Comparison Metrics: View predicted prices with percentage changes from current values
• Session State Persistence: Results remain visible across page interactions

CRUD Operations (Prediction History Management)

• CREATE: Save prediction results with optional notes to JSON database
• READ: View all saved predictions in a searchable table format
• UPDATE: Edit notes on existing predictions with timestamp tracking
• DELETE: Remove individual predictions or clear entire database
• EXPORT: Download prediction history as JSON for backup/analysis
• Database: File-based JSON storage in data/predictions_history.json

Quick Setup (Windows PowerShell)

1. Create and activate a virtual environment

   python -m venv .venv
   .\.venv\Scripts\Activate.ps1

2. Install dependencies

   pip install -r requirements.txt

3. Run the Streamlit app locally

   streamlit run src\stratasight.py

   Or, to use the virtual environment's Python explicitly:

   & ".\.venv\Scripts\python.exe" -m streamlit run src\stratasight.py

The app will open in your browser at http://localhost:8501

How to Use

Generate Predictions Tab

1. Select a Stock: Choose from TSLA, GME, AMD, or AAPL
2. Set Forecast Horizon: Pick how many days ahead to predict (7-90 days)
3. Choose Models: Select LSTM, Prophet, or both for comparison
4. Generate Predictions: Click the button and view results with interactive charts
5. Save Results (optional): Expand "Save This Prediction" to store results with notes

Prediction History (CRUD) Tab

• View History: Browse all saved predictions with timestamps and details
• Update Notes: Select a prediction ID and edit its notes
• Delete Records: Remove individual predictions or clear all at once
• Export Data: Download prediction history as JSON backup

Technical Notes

• Prophet Installation: Can be system-dependent. See Prophet documentation for OS-specific requirements
• Model Storage: Trained models are cached in models/saved/ directory
  • LSTM models: .keras format
  • Prophet models: .pkl format (joblib)
• Data Storage: Prediction history saved in data/predictions_history.json
  • JSON format for human-readable storage
  • Includes ID, timestamp, ticker, prices, predictions, and notes
  • No external database required
• Data Normalization: LSTM uses MinMaxScaler for consistent training and prediction
• Session State: Streamlit session state preserves predictions across page interactions
• Architecture:
  • LSTM: 2-layer stacked LSTM with dropout regularization
  • Prophet: Handles trends, seasonality, and outliers automatically

Deployment

For deployment to Streamlit Cloud:

1. Push your repository to GitHub
2. Connect to Streamlit Cloud
3. Point the app to src/stratasight.py
4. The included requirements.txt will handle dependencies

Project Structure

StrataSight/
├── src/
│   ├── stratasight.py           # Main Streamlit app with CRUD
│   ├── data_fetch.py             # yfinance data fetching
│   └── models/
│       ├── lstm_model.py         # LSTM training and inference
│       └── prophet_model.py      # Prophet training and inference
├── models/
│   └── saved/                    # Cached trained models (.keras, .pkl)
├── data/
│   └── predictions_history.json  # CRUD database (auto-created)
├── requirements.txt              # Python dependencies
├── CRUD_DOCUMENTATION.md         # Technical CRUD documentation
├── CRUD_USER_GUIDE.md            # User guide for CRUD features
└── README.md