Implementation of two causal methods based on predictions in reconstructed state spaces.

Area of Science:

• Complex Systems
• Nonlinear Dynamics
• Causal Inference

Background:

• Deterministic dynamics in data can reveal causal relationships.
• Reconstructed state space methods are used for causal analysis.
• Existing methods face limitations in complex systems.

Purpose of the Study:

• Introduce novel algorithms for causal relationship detection.
• Analyze causality in bivariate and potentially multivariate time series.
• Address limitations of current state-space approaches.

Main Methods:

• Developed two algorithms: cross-prediction and predictability improvement.
• Applied methods to time series data with dominant deterministic dynamics.
• Investigated performance and reliability in various scenarios.

Main Results:

• Cross-prediction method is faster and has fewer false negatives.
• Predictability improvement method aids causal detection and identifies key observables for prediction enhancement.
• Identified weak observability due to complex nonlinear dynamics as a cause for method unreliability.

Conclusions:

• The proposed algorithms effectively detect causality in deterministic systems.
• Method reliability is linked to data observability, not inherent flaws.
• Findings offer new perspectives on causality detection in complex systems.