Reverse engineering for causal discovery based on monotonic characteristic of causal structure

Bayesian networks provide a useful tool for causal reasoning among random variables (nodes) in fields. However, a critical limitation is that it is extremely difficult to obtain an effective causal structure from datasets. In-depth research is required to design a more sophisticated learning algorithm, despite various such algorithms having been introduced to date. In this paper, we present a novel learning algorithm based on the monotonic characteristic of causal relations among a subset of nodes. For example, the magnitude of the causality (dependency) between a child node and its parent node is greater than that between a child node and its grandparent node. Therefore, a child node obtains a higher causality score under its parent node than its grandparent node. We identified the monotonic characteristic in various datasets and designed the proposed method in order to infer causal relations based on the monotonic characteristic. Our experimental results demonstrate that the proposed method significantly improves performance compared to previous methods.