Abstract
This study proposes Evolutionary Causal Discovery (ECD) for causal discoverythat tailors response variables, predictor variables, and correspondingoperators to research datasets. Utilizing genetic programming for variablerelationship parsing, the method proceeds with the Relative ImpactStratification (RIS) algorithm to assess the relative impact of predictorvariables on the response variable, facilitating expression simplification andenhancing the interpretability of variable relationships. ECD proposes anexpression tree to visualize the RIS results, offering a differentiateddepiction of unknown causal relationships compared to conventional causaldiscovery. The ECD method represents an evolution and augmentation of existingcausal discovery methods, providing an interpretable approach for analyzingvariable relationships in complex systems, particularly in healthcare settingswith Electronic Health Record (EHR) data. Experiments on both synthetic andreal-world EHR datasets demonstrate the efficacy of ECD in uncovering patternsand mechanisms among variables, maintaining high accuracy and stability acrossdifferent noise levels. On the real-world EHR dataset, ECD reveals theintricate relationships between the response variable and other predictivevariables, aligning with the results of structural equation modeling andshapley additive explanations analyses.