Predicting the Effect of Fly Ash Dosage on the Compressive Strength of Various Concretes using Machine Learning (ML) Techniques

Abstract

Fly Ash (FA) is a widely used pozzolanic supplement in the production of sustainable concretes in the construction industry. FA is incorporated as a partial replacement of cement in various types of concrete, presenting evident advantages, such as reducing CO₂ emissions, cost-efficient mixes, improved life, and strength parameters of concrete. In response to growing environmental challenges, the development of machine learning (ML) predictive models is a time-consuming process. So, the current study is focusing on developing ML-based models to predict the compressive strength of fly ash-based concretes. The employed modeling techniques include Extra Trees (ET), Random Forest (RF), and XGBoost (XGB). A broad and credible dataset of 543 compressive strength values was compiled from existing literature sources. The nine influential parameters include Mix design proportions (cement, fine aggregate, coarse aggregate), type of concrete, cement type, admixtures, Fly Ash, water-to-binder, temperature, curing days, and relative humidity, which were considered significant input features. The performance of the models was assessed using various statistical parameters including Mean Square Error (MSE), Mean Absolute Error (MAE), Root Mean Squared Log Error (RMSLE), and the coefficient of determination (R²). All individual and ensemble models are based on Extra Trees (ET), Random Forest (RF), and XGBoost (XGB). Extra Trees has demonstrated reliability with high-accuracy predictive results with MSE is 66.62 and R² is 0.90. Statistical comparisons indicated that ET models surpassed RF and XGB for predicting compressive strength values.