A Time Series Forecasting Approach Based on Nonlinear Spiking Neural Systems.

Area of Science:

• Computational neuroscience
• Artificial intelligence
• Time series analysis

Background:

• Nonlinear Spiking Neural (NSNP) systems are a theoretical model inspired by biological neurons.
• These systems possess a nonlinear structure, enabling the description of complex dynamic systems.
• Existing forecasting methods may not fully capture the nonlinear dynamics inherent in many time series.

Purpose of the Study:

• To develop a novel time series forecasting approach utilizing Nonlinear Spiking Neural (NSNP) systems.
• To demonstrate the effectiveness of NSNP systems in predicting future values of time series data.
• To compare the proposed NSNP-based method against state-of-the-art forecasting techniques.

Main Methods:

• Time series data is transformed into the frequency domain using redundant wavelet transform.
• A Nonlinear Spiking Neural (NSNP) system is automatically constructed and adaptively trained in the frequency domain.
• The trained NSNP system generates future sequence data for prediction.

Main Results:

• The proposed NSNP-based forecasting approach was evaluated on eight benchmark and two real-life time series datasets.
• Performance was compared against several state-of-the-art forecasting methods.
• Results indicate the availability and effectiveness of the developed NSNP forecasting approach.

Conclusions:

• The novel time series forecasting method based on Nonlinear Spiking Neural (NSNP) systems is effective.
• The frequency domain analysis and adaptive training of NSNP systems contribute to accurate predictions.
• This approach offers a promising alternative for nonlinear time series forecasting.