Entropy-Based Discovery of Summary Causal Graphs in Time Series

Area of Science:

• Causal inference
• Time series analysis
• Information theory

Background:

• Learning causal relationships from time series data is challenging, especially with varying sampling rates.
• Existing methods may struggle to accurately represent complex temporal dependencies.

Purpose of the Study:

• To develop a robust method for constructing summary causal graphs from time series with heterogeneous sampling rates.
• To introduce a novel causal temporal mutual information measure.

Main Methods:

• Propose a new causal temporal mutual information measure for time series.
• Relate this measure to an entropy reduction principle, a variant of the probability raising principle.
• Integrate these concepts into PC-like and FCI-like algorithms for causal graph construction.

Main Results:

• The proposed algorithms effectively construct summary causal graphs from time series data.
• Evaluation on multiple datasets demonstrates the efficacy and efficiency of the developed methods.

Conclusions:

• The novel causal temporal mutual information measure and entropy reduction principle provide a powerful framework for causal discovery in time series.
• The developed algorithms offer an efficient and effective solution for learning causal graphs from heterogeneous time series data.