Stock Movement Prediction using CNN

Overview

This project is a web application that uses a Convolutional Neural Network (CNN) to predict stock price movements (Up, Down, or Flat) based on uploaded stock chart images. The application is built using Streamlit and TensorFlow/Keras, with a Jupyter Notebook (stock_prediction_app.ipynb) containing the implementation.

Features

• Upload stock chart images (JPG, PNG, JPEG) for analysis.
• Predict stock price movement (Up, Down, or Flat) using a pre-trained CNN model.
• Display training and test accuracy of the model.
• Visualize the distribution of predicted classes for test data using a histogram.

Requirements

• Python 3.8+
• Streamlit
• TensorFlow
• Keras
• NumPy
• Matplotlib
• Scikit-learn
• Jupyter Notebook (optional, for running the notebook directly)

Installation

1. Clone the repository:

   git clone <repository_url>
   cd <repository_directory>

2. Install the required packages:

   pip install -r requirements.txt

3. Ensure you have a dataset of stock chart images or modify the load_real_data function to load your own dataset.

Usage

1. Run the Streamlit app:

   streamlit run stock_prediction_app.py

   Note: You may need to convert the Jupyter Notebook (stock_prediction_app.ipynb) to a Python script (stock_prediction_app.py) using a tool like jupytext or by manually copying the code into a .py file.
2. Open the provided URL in your browser (typically http://localhost:8501).
3. Upload a stock chart image to predict the stock movement (Up, Down, or Flat).
4. View the prediction and the class distribution histogram for test data.

File Structure

• stock_prediction_app.ipynb: Jupyter Notebook containing the main application code.
• requirements.txt: List of required Python packages.
• temp_image.png: Temporary file created when uploading an image (generated at runtime).

Model Details

• Architecture: The CNN model consists of three convolutional layers with ReLU activation, each followed by max-pooling, a flatten layer, a dense layer with 256 units, a dropout layer (0.5), and a final dense layer with softmax activation for three classes (Up, Down, Flat).
• Input: Images are resized to 224x224 pixels and normalized to [0, 1].
• Training: The model is trained on a dummy dataset (replace with real stock chart images). It uses the Adam optimizer and categorical cross-entropy loss.
• Output: Predictions are classified into three categories: Up (0), Down (1), or Flat (2).

Notes

• The load_real_data function currently generates random data for demonstration. Replace it with a function to load actual stock chart images and corresponding labels.
• Ensure the uploaded images are valid stock charts for accurate predictions.
• The model is trained for 10 epochs with a batch size of 32. Adjust these parameters as needed for your dataset.
• The application saves uploaded images temporarily as temp_image.png. Ensure write permissions in the working directory.
• To run the notebook directly, use Jupyter Notebook or Jupyter Lab. To run as a Streamlit app, convert to a .py file.