A stock market trading framework based on deep learning architectures.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Financial Market Analysis

Background:

• Market prediction is crucial for financial professionals.
• Machine learning and AI, especially deep learning, are increasingly used for market prediction.
• Existing models often explore feature extraction and time series forecasting separately, leaving room for integrated frameworks.

Purpose of the Study:

• To propose and evaluate a stacked framework combining Convolutional Neural Network (CNN) and Long-Short Term Memory (LSTM) for Nifty 50 stock price prediction.
• To investigate the efficacy of a rich feature set, including diverse financial data, for enhancing prediction accuracy.
• To compare the performance of the proposed framework against traditional investment strategies.

Main Methods:

• Developed a hybrid CNN-LSTM model for time series forecasting.
• Utilized a 20-day look-up period for predicting the next day's closing price.
• Incorporated a comprehensive feature set: Nifty 50 data, foreign indices, technical indicators, currency exchange rates, and commodity prices.

Main Results:

• The CNN-LSTM framework achieved a Mean Absolute Percentage Error (MAPE) of 2.54% over 10 years of data.
• The model successfully captured market movements by integrating diverse financial indicators.
• Demonstrated significant improvement in returns compared to the traditional buy-and-hold strategy.

Conclusions:

• The proposed CNN-LSTM framework offers a robust and effective approach to stock market prediction.
• Integrating diverse financial data enhances the predictive power of deep learning models.
• This advanced framework provides a superior alternative to traditional investment methods for market prediction.