Clustering-enhanced stock price prediction using deep learning.

Area of Science:

• Artificial Intelligence
• Financial Time Series Analysis
• Deep Learning

Background:

• Financial time series prediction is crucial for market analysis.
• Existing deep learning models require optimization for stock price forecasting accuracy.
• Clustering can improve the quality of training data for predictive models.

Purpose of the Study:

• To propose a clustering-enhanced deep learning framework for optimizing stock price prediction.
• To introduce a new similarity measure, Logistic Weighted Dynamic Time Warping (LWDTW), for effective clustering.
• To evaluate the framework's performance using established deep learning models like LSTM, RNN, and GRU.

Main Methods:

• Developed a clustering pre-processing step for time series forecasting.
• Proposed Logistic Weighted Dynamic Time Warping (LWDTW) by extending Weighted Dynamic Time Warping (WDTW).
• Modified the WDTW cost weight function using logistic probability density distribution based on stock return empirical distributions.
• Implemented a clustering-based forecasting framework with LSTM, RNN, and GRU models.

Main Results:

• The proposed clustering-enhanced framework demonstrated excellent forecasting performance on daily US stock price data.
• The combination of LWDTW clustering and the LSTM model yielded the best results across five evaluation metrics.
• LWDTW effectively captures the relative importance of return observations in distance calculations.

Conclusions:

• The proposed clustering-enhanced deep learning framework significantly improves stock price prediction accuracy.
• LWDTW is an effective similarity measure for financial time series clustering.
• The LSTM model integrated with LWDTW clustering offers a superior approach for stock market forecasting.