Enhanced Interpretability in Root Cause Analysis Using Structure-Template Symbolic Regression

Symbolic regression (SR) is a powerful method for creating interpretable models, but it often lacks functions to incorporate prior knowledge and to support root cause analysis. To address this problem, we propose Structure-Template Regression (STR), a method to integrate known structural components into symbolic regression. This allows for the generation of models that not only fit data well, but also reflect the desired behavior of the underlying system. We evaluated STR on two different problems, a hydrodynamic fluid problem of interconnected communicating vessels and a production logistic use case with multiple assembly lines, and show that STR is capable of generating accurate, interpretable models that adhere to integrated structures. Furthermore, we demonstrate that STR facilitates root cause analysis by enabling the tracing of error pathways within the model structure.