ReTimeCausal: Causal Discovery for Irregular Time Series with Consistency Guarantees

Researchers have introduced ReTimeCausal, a novel framework designed to address the challenges of causal discovery in irregularly sampled time series, a critical issue in fields like finance, healthcare, and climate science. The primary difficulty lies in the interdependence between missing data imputation and causal structure recovery, where errors in one process can negatively reinforce the other. Unlike existing methods that either separate these tasks or lack explicit consistency mechanisms, ReTimeCausal employs an Expectation-Maximization (EM) based approach. It alternates between imputation and structure learning, utilizing kernel-based sparse regression and structural constraints to ensure mutual consistency throughout optimization. The framework offers theoretical guarantees for structure recovery, extending classical results to scenarios with high missingness and irregular sampling. Experimental evaluations on both synthetic and real-world datasets demonstrate that ReTimeCausal outperforms current state-of-the-art methods, providing more accurate causal graphs under challenging data conditions. This advancement promises improved reliability for risk-sensitive applications where data integrity is often compromised by inconsistent sampling frequencies.