Hybrid preprocessing for neural network-based stock price prediction.

Area of Science:

• Quantitative Finance
• Time Series Analysis
• Machine Learning

Background:

• Multivariate time series data in stock price prediction presents complex interdependencies, challenging accurate forecasting.
• Traditional methods often struggle to capture intricate patterns within financial time series data.

Purpose of the Study:

• To introduce a novel hybrid preprocessing technique for enhancing stock price prediction accuracy.
• To address the challenges posed by intricate interdependencies in multivariate time series data.

Main Methods:

• Empirical Wavelet Transform (EWT) for extracting low and high-frequency components.
• Dynamic Time Warping (DTW) and Differential Dynamic Time Warping (DDTW) for component similarity measurement.
• Sliding windows and Principal Component Analysis (PCA) for high-frequency components, and PCA for low-frequency components.
• Integration of these preprocessed components into neural network models.

Main Results:

• A hybrid preprocessing technique combining EWT, DTW, DDTW, and PCA was developed.
• The proposed method demonstrated a substantial 30% improvement in stock price prediction accuracy.
• Identified correlated patterns within stock price series through component similarity analysis.

Conclusions:

• The hybrid preprocessing method significantly enhances stock price prediction accuracy.
• This approach offers valuable insights for financial market analysis and forecasting.
• The technique shows strong potential for improving the performance of neural network models in financial applications.