A fusion of neural, genetic and ensemble machine learning approaches for enhancing the engineering predictive capabilities of lightweight foamed reinforced concrete beam

Abstract

This research explores lightweight foamed reinforced concrete beams, crucial in modern construction for their strength and reduced weight. It introduces a novel approach, integrating three machine learning models: Neural Networks (NNs), Genetic Algorithms (GAs), and Ensemble Techniques, especially Gradient Boosting Machines (GBM). The study evaluates a dataset of 100 tests under various stress conditions, leveraging NNs for deep learning, GAs for feature optimization, and the robustness of GBM. The results demonstrate NNs achieving 88.5% deflection accuracy, 87% load-bearing capacity, and 86% failure point accuracy. GAs show slightly lower performance, while GBM excels with 90.2%, 91%, and 89% in these areas, respectively. Notably, the combined model significantly improves accuracy, reaching 96.8% in deflection, 97.2% in load-bearing capacity, and 96.5% in failure point prediction. This fusion of diverse machine learning approaches marks a significant advancement in structural engineering, enhancing predictive modeling for concrete beams.