Stock Price Range Forecast

An end-to-end Deep Learning & Machine Learning pipeline that forecasts Price Intervals (Confidence Bands) for any stock. Unlike traditional models that predict a single "wrong" price, this model predicts a Probabilistic Range (90% Confidence Interval) to help traders manage risk.

It features a Triple-Engine Architecture, giving you the ultimate flexibility in forecasting power.

Triple Models

The dashboard allows you to toggle between three powerful forecasting modes:

1. ** LightGBM (Statistical):** Fast, stable, and excellent for volatility-based predictions.
2. ** LSTM (Deep Learning):** A Recurrent Neural Network designed to catch non-linear sequential patterns.
3. ** Ensemble (Hybrid):** The "Gold Standard" mode. Trains both models simultaneously and averages their outputs. This reduces individual model errors and typically provides the most robust forecast.

Key Features

• Probabilistic Forecasting: Predicts the 5th and 95th percentile of future price action.
• Multi-Horizon Support: Forecasts for 5 Days (Swing), 21 Days (Monthly), and 60 Days (Quarterly).
• Global Market Support:
  • ** Indian Stocks:** NSE/BSE tickers (e.g., INFY.NS) with ₹ formatting and India VIX integration.
  • ** US Stocks:** NYSE/NASDAQ tickers (e.g., AAPL) with $ formatting and CBOE VIX.
• Dynamic Calibration: Uncertainty bands automatically widen during high volatility (High VIX/ATR).
• Full Automation: GitHub Actions workflow retrains models every night at market close.

Tech Stack

• Core: Python 3.9+
• ML/DL Engines:
  • LightGBM (Gradient Boosting)
  • TensorFlow/Keras (Deep Learning/LSTM)
• Data Source: yfinance (Yahoo Finance API)
• Visualization: Plotly (Interactive Charts)
• Deployment: Streamlit Cloud (Frontend) & GitHub Pages (Static Reports)

How It Works

The model avoids "predicting the exact future price." Instead, it answers:

1. Lower Bound: "There is a 95% chance the price will stay above this line." (Support)
2. Upper Bound: "There is a 95% chance the price will stay below this line." (Resistance)

Trading Logic (Mean Reversion):

• BUY: When Price dips below the Lower Bound (Statistical Oversold).
• SELL: When Price spikes above the Upper Bound (Statistical Overbought).

Local Installation

To run this project on your own machine:

1. Clone the repo:

   git clone [https://github.com/SameerGadge/Stock-Price-Range-Forecast.git](https://github.com/SameerGadge/Stock-Price-Range-Forecast.git)
   cd Stock-Price-Range-Forecast

2. Install dependencies:

   pip install -r requirements.txt

3. Run the App:

   streamlit run app.py

License

MIT License - feel free to use this for your own trading or research!