📈 Stock Price Prediction using Keras and Other Machine Learning Models

🔍 Overview

This project focuses on predicting future stock prices of publicly traded companies using four popular and effective time series forecasting models:

• Keras LSTM (Long Short-Term Memory)
• Facebook Prophet
• Auto ARIMA
• Hybrid Prophet (combination of Prophet, ARIMA)

These models are implemented using Python and are made accessible through a user-friendly web interface built with Streamlit. Users can enter stock ticker symbols (e.g., AAPL, TSLA) to visualize historical data, forecast future prices, and compare the performance of each model.

---

📁 Files and Folder Structure

• webApp.py: Main Streamlit web app integrating all models.
• App.py: An alternative Streamlit app, possibly with different styling or features.
• aarima.py: Streamlit app focused on Auto ARIMA and Hybrid Prophet models.

---

📚 Libraries and Frameworks Used

• Keras - Deep learning model (LSTM)
• Prophet - Time series forecasting library from Facebook
• pmdarima (Auto ARIMA) - Automatic ARIMA modeling
• Hybrid Prophet - Combines Prophet with other models (ARIMA, LSTM)
• Streamlit - Web interface framework
• Pandas & NumPy - Data handling and numerical computation
• Matplotlib & Plotly - Data visualization
• yfinance - Fetch historical stock data from Yahoo Finance

---

⚙️ How It Works

1. Data Collection

• Users input a stock symbol (e.g., MSFT).
• Historical stock data is downloaded using the yfinance API.

2. Data Preprocessing

• Missing values handled
• Time series formatting (e.g., date conversion, sliding windows for LSTM)
• Scaling of data for LSTM models

3. Forecasting Models

✅ Keras LSTM

• Architecture:
  • Input shape: (1, 100)
  • LSTM layer with 50 units
  • Dense layer with 1 unit (linear activation)
• Trained with adam optimizer and mean_squared_error loss

✅ Facebook Prophet

• Growth: linear
• Seasonality: additive
• Fit using fit() and predict() methods

✅ Auto ARIMA

• Automatically chooses p, d, q parameters
• Trained using fit() and forecasted using predict()

✅ Hybrid Prophet

• Combines Prophet + ARIMA (+ optional LSTM)
• Balances strengths of individual models

4. Evaluation

Models are evaluated using:

• Mean Absolute Error (MAE)
• Mean Squared Error (MSE)

5. Streamlit Web Interface

• Input field for stock ticker
• Forecast plot from all models
• Metrics display (MAE, MSE)
• Optional data download/export

---

📈 Example Use Cases

• Forecast stock prices for investment decisions
• Educational tool for time series model comparison
• Research and development in stock prediction using hybrid models

---

🚀 Future Work

• Hyperparameter tuning for improved accuracy
• Feature engineering (technical indicators, volume)
• Sentiment analysis from financial news or social media
• Model stacking and ensemble learning
• Real-time prediction updates
• Enhanced dashboards and visual reports

---

👨‍💻 Author

Sabyasachi Ghosh
📧 sabyasachighosh008@gmail.com
📞 +91 75015 39881

---

⭐ Feel free to fork, star, and contribute to this repository!