Hyper ensembled extreme gradient boosting techniques via novel triple algorithms to appraise the hardened properties of modern ternary admixed concrete

Abstract

Mechanical parameters used in many design codes can be achieved by expensive and time-consuming experiments or by non-destructive approaches such as estimative modelling. This investigation proposed Extreme Gradient Boost (XGB) for estimating the slump (SL) and compressive strength (CS) of high-performance concrete (HPC). In addition, to bring the results of the models closer to the experimental data and increase the accuracy, algorithms were combined with the model, including Sunflower Optimizer (SFO) and Jellyfish Search Optimize (JSO). The relevant models have been examined in three frameworks: individual, hybrid, and ensemble-hybrid. For this purpose, several evaluators were provided to determine the errors, compare, and accuracy of the presented models. The XGFJ model has demonstrated exceptional performance, achieving remarkable results in terms of RMSE (Root Mean Square Error) and R2 (R-squared) values. Specifically, it has attained an exceptionally small RMSE value of 1.785 for CS and 5.183 for SL, indicating the model’s high precision in predicting these parameters. Additionally, it has achieved the biggest R2 values of 0.9960 for CS and 0.9949 for SL. Additionally, it is worth noting that the XGSF model closely matches the performance of the ensemble form of XGFJ, as evident from its R2 values of 0.9956 for CS and 0.9934 for SL. Based on the study, it was observed that using machine learning to anticipate the mechanical characteristics of concrete is valuable and efficient and can be considered an alternative method instead of time-consuming laboratory methods. This research addresses challenges in predicting HPC properties fueled by the need to overcome drawbacks in traditional methods. Costly and time-intensive laboratory experiments prompted the exploration of alternatives, leading to the proposal of XGB combined with optimization algorithms (SFO and JSO). The study aims to enhance prediction accuracy while tackling broader concerns such as construction costs, material efficiency, and environmental impact. The resource-intensive nature of conventional methods, along with inaccuracies due to material variations, serves as a primary challenge. The proposed resolution advocates for a paradigm shift to machine learning, exemplified by the XGFJ model, showcasing exceptional precision and efficiency in predicting HPC properties.