Prediction of concrete compressive strength using a Deepforest-based model

Abstract

Abstract Concrete compressive strength testing is crucial for construction quality control. The traditional methods are both time-consuming and labor-intensive, while machine learning has been proven effective in predicting the compressive strength of concrete. However, current machine learning-based algorithms lack a thorough comparison among various models, and researchers have yet to identify the optimal predictor for concrete compressive strength. In this study, we developed 12 distinct machine learning-based regressors to conduct a thorough comparison and to identify the optimal model. To study the correlation between compressive strength and various factors, we conducted a comprehensive analysis and selected blast furnace slag, superplasticizer, age, cement, and water as the optimized factor subset. Based on this foundation, grid search and fivefold cross-validation were employed to establish the hyperparameters for each model. The results indicate that the Deepforest-based model demonstrates superior performance compared to the 12 models. For a more comprehensive evaluation of the model’s performance, we compared its performance with state-of-the-art models using the same independent testing dataset. The results demonstrate that our model achieving the highest performance (R2 of 0.91), indicating its accurate prediction capability for concrete compressive strength.

Keywords

• Concrete
• Compressive strength
• Factor
• Machine learning
• Deepforest