A Novel XGBoost and RF-Based Metaheuristic Models for Concrete Compression Strength

Concrete is one of the most utilized substances on the planet. However, it is linked to environmental concerns. Recent researches have sought to resolve the issue by substituting cement with substitute cementitious materials. Due to concrete’s complex composition and large number of input variables, it remains a difficult task to forecast its compression strength. On the one hand, laboratory testing is costly, labor-intensive, and time-consuming; meanwhile, empirical models have been shown to have poor performance. The article proposes Extreme gradient boosting (XGBoost) machine and Random Forest (RF) machine learning-based prediction model to predict the compressive strength of concrete. The model is developed and validated using the dataset of 144 compressive test results of fly ash and silica fume concrete. The best XGBoost model is simulated using 100 maximum depths, 0.3 learning rate, 10 number iterations, and 123 random seats. The best performance of the Random Forest model is achieved for five numbers of iterations, 1000 maximum depth of the tree, and 120 random numbers of seed used. The XGBoost model outperforms (coefficient of correlation of 0.9998 in training and 0.957 in testing) Random Forest model (coefficient of correlation 0.96 in training and 0.947 in testing); however, both the models are concluded to have robust performance. The performance of the model is further confirmed using various statistical performance parameters and Taylor diagram. The paper will propose machine learning-based robust alternatives for prediction of compressive strength of concrete. Since concrete is a highly heterogeneous mixture and its properties vary from place to place, the stimulated models need to validate a wide variety of data.