Automated ensemble learning for shear strength prediction in marine reinforced concrete slabs : a self-optimizing stacked framework

Abstract

Reinforced concrete (RC) slabs are fundamental structural elements widely employed in both general and marine engineering applications, where they are required to ensure load distribution and in-plane stability under various environmental and mechanical stresses. In marine environments, these slabs are particularly vulnerable to degradation mechanisms such as chloride-induced corrosion and cyclic wave loading, which significantly increase the risk of brittle punching shear failures. Such failures, typically resulting from unbalanced shear forces and insufficient reinforcement detailing, can lead to abrupt and catastrophic structural collapse. Existing empirical approaches for predicting punching shear strength often lack the capacity to account for the complex, nonlinear interactions among geometric, material, and environmental parameters, thereby limiting their reliability in marine contexts. To address these limitations, this study proposes an automated machine learning (AutoML) framework that leverages data-driven modeling for accurate and efficient prediction of punching shear strength in RC slabs. The framework integrates automated model selection, hyperparameter optimization, and feature selection within a self-optimizing multi-layer stacking ensemble, eliminating the need for manual intervention and enhancing predictive performance. Model interpretability is achieved through SHapley Additive exPlanations (SHAP) analysis, which identifies the most influential predictors and provides insights into underlying structural behaviors. The proposed methodology is validated using an extensive dataset comprising RC slabs with varying reinforcement configurations, material properties, and exposure conditions, including those specific to marine environments. Results demonstrate superior generalization capability and improved prediction accuracy compared to traditional approaches, highlighting the framework’s potential as a practical tool for structural design, assessment, and durability analysis of RC elements in marine infrastructure systems.