Intelligent Models for Prediction of Compressive Strength of Geopolymer Pervious Concrete Hybridized with Agro-Industrial and Construction-Demolition Wastes

Abstract

In modern civil engineering, precisely predicting the mechanical properties of waste-modified geopolymer concrete is a vital challenge. Machine learning (ML) offers a powerful tool for such predictive analysis. This article presents an experimental and python-based intelligent ML modeling study on a type of geopolymer (GP) pervious concretes developed using agro-industrial waste products. The slag-based composite mixes were developed with the varying dosages of agro-waste, i.e., sugarcane bagasse ash (0 to 20% by weight of slag) and construction and demolition waste in the form of recycled coarse aggregates (0 to 100% by weight of natural aggregates). The aqueous solution of liquid Na2SiO3 and NaOH pellets were used as an alkali activator solution. A total of 13 different mix proportion designs were developed, and for every individual sample mix, the results were obtained from laboratory tests. The ML analysis was carried out to compute the compressive strength by applying following models: Multiple Linear Regression, tuned Gradient Boost, AdaBoost, and XGBoost Regressions. Further, an ensemble technique that combines the predictions from multiple ML algorithms together to make more accurate predictions than any individual model was also developed for a more accurate and robust prediction through the “Voting Regressor” technique. From the analysis of the obtained results, the ML models associated with Ada Boost tuned performed better. As the ensemble voting regressor models were given higher weightage, these regressors gave the best performance metrics, with lower error rate compared to the independent models.