Core Concepts
The author proposes a two-phase multi-split causal ensemble model to combine different causality base algorithms, aiming to improve robustness and reliability in causal inference.
Abstract
The content discusses a novel data-driven two-phase multi-split causal ensemble model for time series. It combines various causal inference methods, evaluates the trustworthiness of results, and optimizes the final causal strength matrix by removing indirect links.
Reviewing the key points:
Introduction to causal inference in time series data.
Explanation of Granger Causality Test, Transfer Entropy, PCMCI+, and Convergent Cross Mapping.
Proposal of a two-phase ensemble model combining different algorithms.
Data partitioning and GMM ensemble phase for processing results.
Rule ensemble phase with three rules for integrating intermediate results.
Model optimization to remove indirect causal links from the final result.
Stats
The GC has been widely used for causal inference in time series analysis since its introduction in 1969 [22].
Transfer entropy (TE) is capable of detecting non-linear causal relationships [5].
PCMCI+ extends PCMCI by detecting contemporaneous links [21].
Convergent Cross Mapping (CCM) is based on the theory of non-linear state space reconstruction [6].