Machine learning approach for predicting the compressive and flexural strengths of fiber-reinforced concrete mixtures

Abstract. This study employed the Extreme Gradient Boosting (XGBoost) Machine Learning (ML) model to predict the compressive and flexural strengths of concrete mixtures reinforced with 0.5, 1.0, and 1.5% of polyvinyl alcohol (PVA) fibers volume fractions at 28 days of concrete age. A significant increase in the compressive and flexural strengths of the resulting concrete mixtures was recorded as a result of the incorporation of PVA fibers, particularly at the higher fiber volume fraction level when compared with that of control mixture produced without the addition of PVA fibers. The ML model-predicted strengths data was found to be in close conformity with the experimentally tested one, pointing at the high accuracy of XGBoost model. Statistical analysis of the output variable data yielded lower error and higher R2 values (0.931 for compressive and 0.922 for flexural strengths) which confirm the suitability of the XGBoost model to predict the two types of strength characteristics of the PVC fiber reinforced concrete mixtures with high accuracy. The use of ML-model to predict the strength characteristics of PVA fiber-reinforced concrete mixtures is viewed to be a time-efficient method having high level of accuracy.