Particle swarm optimization-based extreme gradient boosting for concrete strength prediction

Although concrete is one of the most widely used construction materials in civil engineering, accurate prediction of concrete strength is very difficult for its highly non-linearity of age, concrete components, and physical and chemical reaction process. To address this problem, we proposed an improved extreme gradient boosting (XGBoost) for concrete strength prediction based on parameter optimization. By aggregating multiple `weak' prediction models (e.g., decision trees), XGBoost is able to build a prediction model with high execution speed and model performance. To further improve the performance of XGBoost, comprehensive learning particle swarm optimizer (CLPSO) is employed for parameter selection. Five-fold cross validation is introduced to better evaluate the prediction performance. Experimental results from concrete strength prediction have shown our proposed CLPSO-XGBoost is able to outperform other well-known machine learning models in comparison in terms of robustness and accuracy, which confirms that CLPSO-XGBoost can be a useful tool to predict the compressive strength in practical civil engineering applications.