Reconstructing directional causal networks with random forest: Causality meeting machine learning.

Area of Science:

• Computational Biology
• Network Science
• Machine Learning

Background:

• Traditional methods for detecting causality often focus on pairwise interactions.
• A gap exists between pairwise causality detection and full causal network reconstruction.
• Complex systems require advanced methods for understanding causal relationships.

Purpose of the Study:

• To introduce a novel framework for causal network reconstruction named Importance Causal Analysis (ICA).
• To bridge the gap between existing mutual causality detection techniques and comprehensive causal network building.
• To validate the efficacy of ICA in identifying true causal links within complex networks.

Main Methods:

• Developed the Importance Causal Analysis (ICA) framework inspired by decision tree algorithms.
• Designed ICA as a network-level approach for causal inference.
• Applied ICA to both simulated benchmark systems and real-world datasets.

Main Results:

• ICA demonstrated the ability to reconstruct causal networks effectively.
• The framework successfully identified true causal relations in complex network structures.
• Validation on benchmark and real-world data confirmed ICA's potential.

Conclusions:

• Importance Causal Analysis (ICA) offers a novel and effective approach to causal network reconstruction.
• The method addresses limitations of traditional causality detection by providing a network-level perspective.
• ICA shows promise for uncovering true causal relationships in diverse and complex systems.