Temporal state-space model for forecasting slow-wave EEG power in non-human primates.

Area of Science:

• Neuroscience
• Time Series Analysis
• Signal Processing

Background:

• Accurate forecasting of brain activity is crucial for establishing neural dynamics baselines before interventions.
• Longitudinal electroencephalography (EEG) data presents complex temporal structures requiring advanced analytical methods.

Purpose of the Study:

• To develop and evaluate an individualized time series forecasting framework for predicting slow-wave activity (SWA) power in non-human primates.
• To assess the performance of the proposed framework against traditional forecasting methods.

Main Methods:

• Development of an individualized forecasting framework utilizing Trigonometric, Box-Cox transformation with ARMA errors and Trend and Seasonal/Periodic Components (TBATS).
• Application of the TBATS model to predict SWA power in longitudinal EEG data from non-human primates.
• Comparison of TBATS model performance against naive baseline, Holt-Winters (HW), and Seasonal ARIMA (SARIMA) models.

Main Results:

• The TBATS framework demonstrated comparable or superior out-of-sample accuracy in predicting SWA power compared to baseline and traditional methods.
• TBATS effectively captured subject-specific latent temporal structures in neural dynamics.
• The model proved to be data-efficient and interpretable for forecasting longitudinal EEG data.

Conclusions:

• The TBATS framework is a robust and effective tool for individualized forecasting of neural dynamics, specifically SWA power.
• This approach facilitates the establishment of reliable baselines for neural activity prior to experimental interventions.
• The findings support the utility of TBATS in analyzing complex, longitudinal EEG data in neuroscience research.