Time-Series Analysis

Abstract

Time-series analysis is useful for many applications in the field of epilepsy. From data sources including seizure recording devices, patient seizure diaries, functional MRI and EEG, time-series analysis forms the foundation of inference and prediction in many data streams. This chapter highlights basic foundations behind several classical statistical models in time-series analysis that have been used in epilepsy research to date. We first review foundations of linear time-series models for univariate time-series data, including the autoregressive moving average class of models and conditional heteroskedasticity models. Next, we cover several multivariate time-series models, including vector autoregressive and vector autoregressive moving average models. We discuss incorporation of brain states into time-series models and cover state-space models, including dynamic linear models and hidden Markov models, with discussion of an example of a nonlinear extension via regime-switching models. We end with a brief summary of circular data analysis, which can be used to investigate cycles in time-series data. Examples are provided in the R programming language using a simulated dataset from chronic ambulatory electrocorticography acquired from the responsive neurostimulation system (RNS®).

Topics & Concepts