Discrepancies Between MLP and LSTM Models for Nonlinear Granger Causality

[ZahraMoradi1998]

Hello,

Thank you for sharing this valuable resource.

I am using the nonlincausality package to analyze Granger causality across 8 different brain regions using EEG time series data. I initially tested the code on simulated data where the true Granger causality was known, and the results were accurate. However, when applying both MLP and LSTM models to my real EEG data, I noticed significant differences in the results, which I’m finding difficult to interpret.

Could you please provide some guidance on the following:
What might cause MLP and LSTM models to produce different Granger causality results?
How can I determine which model is more appropriate for analyzing my EEG time series data?
Are there best practices for selecting models in nonlinear Granger causality testing?

Thank you in advance for your help.

Best regards