Shear resistance in high-strength concrete beams without shear reinforcement: A new insight from a structured implementation of Explainable Artificial Intelligence

Abstract

This paper presents a data-driven modeling methodology based on Explainable Artificial Intelligence (XAI) integrated with Genetic Programming (GP), called XAI-GP, to develop a transparent and practical model for predicting the shear strength of High-Strength Concrete (HSC) beams without shear reinforcement. First, three AI models were trained using empirical data from the literature, and the most accurate model was selected. XAI techniques (SHAP and Breakdown explainers) were then applied in a structured manner to identify key input parameters influencing ultimate shear stress, ensuring model robustness and preventing misleading conclusions. Using these insights, a new shear strength expression was formulated via GP, balancing accuracy, safety, and compliance with design standards. The XAI-GP model was evaluated against empirical models from concrete design codes and previous studies, explicitly considering both safety and accuracy. Results demonstrate that XAIGP enhances predictive performance while ensuring usability and trustworthiness for engineers.