Shear capacity prediction for FRCM-strengthened RC beams using Hybrid ReLU-Activated BPNN model

Abstract

This study presents a robust Hybrid ReLU-Activated Backpropagation Neural Network (BPNN) model for predicting shear strength (VFRCM) in RC beams reinforced with Fiber-Reinforced Cementitious Matrix (FRCM) composites. The model demonstrates exceptional accuracy, boasting a coefficient of determination (R2) of 0.924 and a mean root mean squared error (RMSE) of 10.611 kN on the test set. Leveraging k-fold cross-validation, the model's performance is rigorously assessed across ten data splits, emphasizing its reliable predictive capabilities. Hyperparameter optimization, guided by Particle Swarm Optimization (PSO), reveals an optimal value of α = 0.5757, minimizing the Mean Squared Error (MSE) to 74.53 for the dataset. Feature importance analysis underscores the significant role of critical parameters in shear strength prediction, with the longitudinal reinforcement ratio (ρl) emerging as the most influential feature, contributing 37.55 % to the model's predictive power. Additionally, the effective beam depth (d) is identified as the second most influential feature, providing approximately 16 % to the accurate prediction of VFRCM in RC beams strengthened with FRCM composites.