Metaheuristic-based machine learning modeling of the compressive strength of concrete containing waste glass

Abstract

Waste glass (WG) can be used as fine aggregate and powder in concrete mixtures, preventing pollution induced by this non-biodegradable material. The properties of WG-included concrete should be examined before its practical use. Compressive strength (CS) is one of the most crucial characteristics of concrete, and the measurement of which needs time-consuming and expensive experiments. The use of machine learning (ML) methods for modeling the CS of concrete can help achieve more reliable and precise models. In this study, a comprehensive database of WG-included concrete was collected from the literature. Next, four ML methods, including support vector regression (SVR), least-square support vector regression (LSSVR), adaptive neuro-fuzzy inference system (ANFIS), and multilayer perceptron neural network (MLP) were served in the CS modeling. A recently proposed metaheuristic method, called marine predators algorithm (MPA), was proposed to optimize the control parameters of the ML models to guarantee generalized accuracy. Results indicate that the hybrid LSSVR-MPA model outperforms the other developed ML models comparing the error metrics with an RMSE = 2.447 MPa and R2 = 0.983. The sensitivity analysis reveals that replacing the cement with WG powder decreases the CS, whereas serving the WG as the replacement for natural fine aggregate improves the CS.