Epilepsy impacts 65 million globally, necessitating advanced research for effective treatment strategies. This study evaluates the HMM-MVAR model’s utility in epilepsy, particularly in seizure prediction and brain network analysis. Initial simulations revealed its superiority over traditional HMM and HMM-TDE approaches. Utilizing the Helsinki EEG database, the model achieved a notable average error of 2.759 in seizure timing prediction. Further application on sEEG data from TLE patients in the HUP database identified six distinct state-time transition sequences, providing critical insights into epileptic network dynamics. These results underscore the potential of the HMM-MVAR model in enhancing clinical epilepsy management and understanding its underlying mechanisms.