Activity Number:
|
39
- Advances in Time Series: Statistics Meets Machine Learning
|
Type:
|
Invited
|
Date/Time:
|
Sunday, August 7, 2022 : 4:00 PM to 5:50 PM
|
Sponsor:
|
Section on Statistical Computing
|
Abstract #319295
|
|
Title:
|
Spectral Nonlinear Granger Causality for Multivariate Time Series
|
Author(s):
|
Hernando Ombao* and Archishman Biswas
|
Companies:
|
King Abdullah University of Science and Technology and King Abdullah University of Science and Technology
|
Keywords:
|
Spectral analysis;
Non-linear dependence;
Granger-causality;
Network;
Multivariate time series;
Brain signals
|
Abstract:
|
One of the key goals in analyzing multivariate time series is to characterize and estimate the cross-dependence structure among the components. Traditional approaches (e.g., coherence and correlation) capture only linear dependence. This serious limitation could lead to false conclusions under non-linearity. Keeping this as our motivation, we propose a procedure for identifying non-linear and frequency-band specific Granger causality (Spec NLGC) connections. The advantages of the Spec NLGC approach over traditionally used VAR-based models will be demonstrated using simulations and in the analysis of epileptic seizure EEG data. It was able to uncover non-linear dynamics and yielded novel and insightful findings. The time-evolving Spec NLGC connections gives more meaningful insights regarding the frequency specific connectivity changes at the onset of epileptic seizure as compared to VAR based PDC connections. These confirm the viability of the proposed algorithm as a good connectivity exploration tool.
|
Authors who are presenting talks have a * after their name.