A metaheuristic causal discovery method in directed acyclic graphs space

Causal discovery plays a vital role in the human understanding of the world. Searching a directed acyclic graph (DAG) from observed data is one of the most widely used methods. However, in most existing approaches, the global search has poor scalability, and the local search is often insufficient to discover a reliable causal graph. In this paper, we propose a generic metaheuristic method to discover the causal relationship in the DAG itself instead in of any equivalent but indirect substitutes. We first propose several novel heuristic factors to expand the search space and maintain acyclicity. Second, using these factors, we propose a metaheuristic algorithm to further search for the optimal solution closer to real causality in the DAG space. Theoretical studies show the correctness of our proposed method. Extensive experiments are conducted to verify its generalization ability, scalability, and effectiveness on real-world and simulated structures for both discrete and continuous models by comparing it with other state-of-the-art causal solvers. We also compare the performance of our method with that of a state-of-the-art approach on well-known medical data.& COPY; 2023 Elsevier B.V. All rights reserved.

Automated summary

This paper proposes a generic metaheuristic method for causal discovery by directly finding causal relationships in a directed acyclic graph. Several novel heuristic factors are introduced to expand the search space and maintain acyclicity. A metaheuristic algorithm is then used to search for an optimal solution closer to real causality. The proposed method is theoretically proven and extensively validated through experiments comparing it with other state-of-the-art causal solvers on real-world and simulated structures.