Quantum-Enhanced Stock Prediction

A comparative analysis project exploring classical machine learning versus quantum-enhanced prediction models for stock market forecasting using PennyLane. The project includes data ingestion, data preprocessing, model training, and deployment using Docker and Kubernetes.

---

Table of Contents

• Overview
• Features
• Installation
• Usage
  • Data Ingestion
  • Data Preprocessing
  • Classical ML Model
  • Quantum-Enhanced Model
• Deployment
• Future Scope
• References
• Contributing
• License

---

Overview

Stock market prediction is highly challenging due to the complex, non-linear, and high-dimensional nature of financial data. This project evaluates whether integrating quantum computing techniques with classical machine learning can significantly enhance predictive accuracy. Our quantum-enhanced model, built using PennyLane, achieved 94% accuracy compared to 81% for classical models.

---

Features

• Data Ingestion: Fetches historical stock data from the Alpha Vantage API and stores it as CSV files.
• Data Preprocessing: Cleans and enriches data by computing technical indicators (Moving Averages, RSI, MACD, etc.), applying noise reduction techniques (Savitzky–Golay filter), and generating lag features.
• Classical ML Model: Implements a Random Forest Regressor for benchmark predictions.
• Quantum-Enhanced Model: Utilizes variational quantum circuits (with RY rotations and CNOT gates) and the Adam optimizer to achieve superior prediction accuracy.
• Deployment: Implements a Flask API for predictions, containerized with Docker and orchestrated via Kubernetes for scalability.
• Comparative Analysis: Evaluation metrics include RMSE, MAE, MAPE, and overall accuracy.

---

Installation

1. Clone the repository:

   git clone https://github.com/yourusername/quantum-stock-prediction.git
   cd quantum-stock-prediction
   

2. Create a virtual environment:

   python -m venv venv
   source venv/bin/activate   # For Linux or macOS
   venv\Scripts\activate      # For Windows
   

3. Install Dependencies:

   pip install -r requirements.txt
   

4. **Configure your API keys: ** Replace YOUR_API_KEY in data_ingestion.py with your actual Alpha Vantage API key.

Usage

Data Ingestion

• Script: data_ingestion.py
• Purpose: Fetch historical stock data for multiple symbols from the Alpha Vantage API and save the data as CSV files in the data/raw directory.
• How to Run:

  python data_ingestion.py
  
  

Data Preprocessing

• Script: data_preprocessing.py
• Purpose:
  • Load raw stock data from CSV files in the data/raw directory.
  • Apply noise reduction using the Savitzky–Golay filter to smooth out the closing price series.
  • Compute technical indicators such as Moving Averages (MA_10, MA_50, MA_20), RSI, MACD, and Volatility.
  • Generate lag features (e.g., lag_1, lag_2, lag_3) to capture temporal dependencies.
  • Combine data from multiple stock symbols into a single DataFrame and add a symbol identifier.
  • Split the enriched data into training and testing sets while preserving the sequential order.
  • Save the processed dataset and the train-test splits to the data/processed and data/splits directories respectively.
• How to Run:

  python data_preprocessing.py
  

Classical ML Model

• Script: classical_ML.py
• Purpose:
  • Load the preprocessed dataset from data/processed/processed_data.csv.
  • Select key technical indicators (e.g., MA_10, MA_50, RSI, MACD, Signal_Line) as input features.
  • Use the closing price as the target variable.
  • Split the dataset into training and testing sets, maintaining the time-series order.
  • Train a Random Forest Regressor on the training set.
  • Evaluate the model using performance metrics such as RMSE, R², MAE, MAPE, and F1-score.
  • Save the trained model as a .pkl file in the models/classical directory.
  • Generate stock recommendations by simulating future returns and identifying optimal sell dates based on predicted returns.
• How to Run:

  python classical_ML.py
  
  

Quantum-Enhanced Model

• Script: quantum_model.py
• Purpose:
  • Load the preprocessed dataset (features and target) from data/processed/processed_data.csv.
  • Normalize the features and target values to ensure numerical stability in the quantum circuit.
  • Define a variational quantum circuit using PennyLane:
    • Data Encoding: Use RY gates to encode normalized feature values into quantum states (via ( \theta = \arcsin(\text{clip}(x, -1, 1)) )).
    • Variational Layers: Apply 4 layers where each layer performs parameterized RY rotations followed by entanglement using CNOT gates.
    • Measurement: Compute the expectation value of the Pauli-Z operator on the first qubit.
  • Utilize the Adam optimizer to update the circuit’s parameters over 50 epochs, minimizing a cost function that combines Mean Squared Error (MSE) and L2 regularization.
  • Generate predictions on both the test set and the full dataset, then de-normalize the predicted outputs.
  • Evaluate the model performance using metrics such as RMSE, MAE, and MAPE.
  • Save the trained quantum model parameters (weights) in the models/quantum directory and output evaluation metrics to a report file.
• How to Run:

  python quantum_model.py
  
  

Deployment

• Flask Backend:

  • A Flask API is used to serve prediction and recommendation endpoints.
  • This API integrates both the classical and quantum models, handling incoming requests and returning predictions.

• Docker Containerization:

  • The entire application is containerized using Docker to ensure a consistent environment across different systems.
  • A multi-stage Dockerfile is used to separate development dependencies from the runtime environment, resulting in a lean production image.
  • Example command to build the Docker image:

    docker build -t quantum-stock-prediction .

Future Scope

• Real-Time Data Integration:

  • Incorporate live-streaming stock data to enable continuous and dynamic forecasting.

• Advanced Quantum Algorithms:

  • Explore new quantum circuit designs and error mitigation strategies as quantum hardware advances.

• Hybrid Quantum-Classical Models:

  • Further integrate classical and quantum techniques to harness the strengths of both for improved prediction accuracy.

• Expanded Feature Set:

  • Integrate supplementary data sources (e.g., news sentiment, social media trends) to enhance model inputs and predictive power.

• Cross-Industry Applications:

  • Adapt the developed framework for predictive analytics in sectors such as healthcare, manufacturing, logistics, and telecommunications.

• Cloud-Based Deployment:

  • Scale the system via cloud platforms (AWS, GCP) with auto-scaling, monitoring, and improved resource management.

License

This project is licensed under the MIT License. See the LICENSE file for details.

Coded by Uttakarsh

```