What does this research mean for the field?

Ensemble-based machine learning models, specifically Extreme Gradient Boosting, can accurately predict the punching shear capacity of reinforced concrete flat slabs (R² = 0.948) and identify average effective depth as the dominant influencing factor. Novelty: ClaimNovelty.METHODOLOGICAL. Consensus alignment: ConsensusAlignment.NEUTRAL.

What question did this study set out to answer?

The aim is to develop accurate and explainable machine learning models for predicting the punching shear capacity of reinforced concrete flat slabs.

June 12, 2026Open Access

Explainable and comparative machine learning models for punching shear prediction in reinforced concrete flat slabs

QSQaim ShahUniversity of Engineering and Technology Peshawar IUIrfan UllahHohai University WKWaheed Ali KhosoCECOS University

Key Points

The aim is to develop accurate and explainable machine learning models for predicting the punching shear capacity of reinforced concrete flat slabs.
Constructed a database with 378 validated punching shear test results through outlier detection.
Developed and optimized eight machine learning algorithms using extensive hyperparameter tuning.
Evaluated model performance using various statistical metrics and explainable AI techniques.
Extreme Gradient Boosting achieved the highest accuracy (R^2 = 0.948) and lowest prediction errors.
Average effective depth was identified as the primary factor influencing punching shear capacity.
Ensemble models significantly outperformed linear and gradient-based methods in predictive accuracy.

Abstract

Abstract Accurate prediction of punching shear capacity in reinforced concrete flat slabs remains a critical challenge due to the highly nonlinear interaction between geometric, material, and reinforcement parameters. Traditional design formulations often exhibit excessive conservatism, leading to inefficient material use and increased environmental impact. This study presents a comprehensive and explainable machine learning framework for predicting the ultimate punching shear capacity of reinforced concrete flat slabs. An extensive experimental database comprising 378 validated test results was compiled after rigorous outlier detection using unsupervised learning techniques. Eight machine learning algorithms, including linear, instance-based, kernel-based, neural, and ensemble methods, were systematically developed and optimized through rigorous hyperparameter tuning to ensure fair performance comparison. Model performance was evaluated using multiple statistical metrics, revealing that ensemble-based models significantly outperform linear and gradient-based approaches. Extreme Gradient Boosting achieved the highest accuracy, with a coefficient of determination of 0.948 and the lowest prediction errors. To enhance transparency and physical consistency, explainable machine learning techniques, including SHapley Additive exPlanations and partial dependence analysis, were employed to interpret model behaviour and identify governing parameters. The results confirm that average effective depth is the dominant factor influencing punching shear capacity, followed by concrete compressive strength and geometric perimeter characteristics, while reinforcement parameters play a secondary role. A user-friendly graphical interface was further developed to facilitate practical application of the trained models. The proposed framework demonstrates strong potential for improving prediction reliability, supporting performance-based design, and enabling more sustainable and material-efficient reinforced concrete slab design.

اسأل الذكاء الاصطناعي

Bookmark

View Full Paper