Causal Inference from Time Series: Methods for Discovering, Explaining, and Estimating Causal Relationships

Across various fields of engineering and science, there is great interest in studying causal relationships between time series. Distinguishing cause from effect is difficult in practice for many reasons, including limited access to data, unknown functional relationships, and unobserved confounding factors. Due to these challenges, modern causal inference requires methods that can perform robust detection and estimation, quantify uncertainty, and explain how model?s inputs contribute to its predictions. These challenges are further compounded in time series settings, where autocorrelation and temporal patterns can skew inference. This thesis introduces several contributions to the field of causal inference that aim to address these concerns. The first part of the thesis examines approaches to causal discovery and the detection and estimation of causal relationships, with a focus on time-series data. The second part of the thesis considers the explanation of causal models and proposes methods that quantify the strength of causation, detect interactions between multiple causes, and quantify the extent to which a variable contributed to a particular outcome. The proposed analyses are validated through experiments using both synthetic and real datasets.