Causal Discovery Combining K2 with Brain Storm Optimization Algorithm.

Area of Science:

• Causal inference
• Machine learning
• Data science

Background:

• Causal discovery is crucial for scientific discovery and data science.
• Constraint-based causal discovery methods struggle with large, complex datasets.
• Existing methods often fail to scale effectively for large-scale causal structure learning.

Purpose of the Study:

• To develop an improved causal structure learning algorithm for large datasets.
• To enhance the efficiency and scalability of causal discovery methods.
• To address limitations of traditional constraint-based approaches in complex scenarios.

Main Methods:

• Introduced Brain Storm Optimization (BSO) for searching optimal topological node orders.
• Combined K2 algorithm with BSO, creating the K2-BSO algorithm.
• Designed a novel distance function for BSO's clustering step, tailored to K2's mechanism.

Main Results:

• K2-BSO significantly reduces the search space from graph space to topological order space.
• The method demonstrated superior performance compared to traditional search and score methods.
• Experimental results on real-world datasets confirmed K2-BSO's effectiveness over genetic algorithm-based methods.

Conclusions:

• K2-BSO offers a scalable and efficient solution for causal discovery in large datasets.
• The algorithm successfully overcomes the limitations of traditional constraint-based methods.
• This approach holds significant promise for advancing causal inference in complex systems.