A combined transfer learning physics informed interpretable machine learning approach to modelling the shear strength of concrete walls

Data scarcity is a persistent challenge in civil engineering, especially for emerging materials and uncommon structural configurations. This study presents a novel, interpretable machine learning framework combining Transfer Learning (TL) and Physics-Informed Machine Learning (PIML) using Genetic Expression Programming (GEP). Physical knowledge, based on ACI shear strength equations, is injected through feature engineering, while the 2M2P algorithm enables knowledge transfer from common wall geometries to less common ones with limited data. Results show that the TL-PIML models often outperform both traditional data-driven models and existing code equations, particularly under severe data constraints. The approach yields interpretable, physically consistent models that achieve higher utilizing limited training data. This work demonstrates the effectiveness of combining TL and PIML for improving predictive modelling in structural engineering and offers a robust strategy for extending machine learning applications to data-sparse scenarios.