Multivariate time series forecasting method based on nonlinear spiking neural P systems and non-subsampled shearlet...

Area of Science:

• Data Science
• Computational Neuroscience
• Signal Processing

Background:

• Multivariate time series forecasting presents challenges due to nonlinear, non-stationary, and high-dimensional data.
• Existing methods struggle with the complex spatial-temporal dependencies and inter-variable relationships inherent in such data.

Purpose of the Study:

• To develop a novel and effective method for multivariate time series forecasting.
• To address the limitations of current forecasting techniques in handling complex data characteristics.

Main Methods:

• A novel approach combining nonlinear spiking neural P (NSNP) systems with non-subsampled shearlet transform (NSST).
• Multivariate time series are transformed into the NSST domain for analysis.
• NSNP systems are constructed, trained, and used for prediction within the NSST domain, enabling multiscale transform-based prediction.

Main Results:

• The proposed method successfully processes nonlinear and non-stationary time series.
• The multiresolution features of NSST effectively capture inter-variable dependencies.
• Experimental results on five real-life datasets show superior performance compared to state-of-the-art and baseline methods.

Conclusions:

• The proposed NSST-domain NSNP system method is highly effective for multivariate time series forecasting.
• This approach offers a robust solution for complex time series data, outperforming existing techniques.