Assessment of rapid chloride penetrability and cost of high-performance concrete via reliable hybrid and ensemble predictors: a comparative investigation

Reinforced High-Performance Concrete (HPC) is valued for its strength and durability in construction, but it can deteriorate over time due to chloride attacks. The Rapid Chloride Penetrability Test (RCPT) is used to measure concrete's resistance to chloride penetration, but its accuracy is affected by factors like concrete age, test conditions, and specimen size. The study aims to improve the prediction of RCPT and cost for HPC mixtures using machine learning models. It examines the effectiveness of hybrid and ensembled estimation frameworks incorporating Supplementary Cementitious Materials (SCMs) like Fly Ash (FA) and Silica Fume (SF). These SCMs enhance HPC properties and support sustainable construction. This research provides a comparative analysis of various predictive models developed using Radial Basis Function (RBF), Adaptive Boosting (ADA), and several optimization algorithms, including Cheetah Optimization (CHO), Smell Agent Optimization (SAO), and Mountain Gazelle Optimizer (MGO). The study explores the performance of these models individually and in combination. Notably, the ensemble model incorporating ADA with CHO, MGO, and SAO, referred to as the ADA+CHO+MGO+SAO (ACMS) model, demonstrated superior performance. The ACMS model achieved an R2 value of 0.987 for predicting the Rebar Concrete Pullout Test (RCPT) and an impressive 0.993 for cost prediction. Additionally, the model maintained an error rate below 2% for the majority of samples, indicating its high accuracy and reliability. This robust performance underscores the effectiveness of combining these optimization techniques to enhance predictive accuracy in the given context.