A Brief Review on Causal Discovery in Time Series

Temporal data, which captures sequential observations of complex systems over time, is a common data structure generated across diverse fields, including industry, healthcare, and finance. Uncovering the causal relations within such data is highly beneficial for numerous applications. In this review, we present some key concepts, models, and algorithms developed to date for discover causal relationships from observational time series data. To do this, we first introduce the common approach used in causal discovery literature and then explore various methods categorized by approach: Constraint-based methods, Score-based approaches, FCM-based approaches and Granger causality-based approaches. From this review, we conclude that causal discovery in time series data remains an active research area with continual development of new methods across all approach categories, and no single family or method is universally optimal. Each approach depends on assumptions that may or may not hold for specific datasets.

کلیدواژه ها:

Temporal Causal Discovery ، Granger Causality ، Causal Graph ، Temporal Convolutional Networks