A generalized method to predict the compressive strength of high-performance concrete by improved random forest algorithm. (30th November 2019)

• Record Type: Journal Article
• Authors: Han, Qinghua; Gui, Changqing; Xu, Jie; Lacidogna, Giuseppe
• Is Part Of: Construction & building materials. Volume 226(2019)
• Page Start: 734
• Page End: 742
• Publication Date: 2019-11-30
• Subjects: Random forest -- High-performance concrete -- Compressive strength -- Input variable optimization -- Parameter determination; Building materials -- Periodicals; 624.18
• DOI: 10.1016/j.conbuildmat.2019.07.315
• Languages: English
• ISSNs: 0950-0618

Abstract

Highlights: An improved random forest method was proposed to predict HPCCS. Appropriate features for modeling can be obtained by this method. Satisfactory results with default parameter settings can be obtained. It performs well when the input variables in absolute mass form. The prediction accuracy is superior to that of other methods. Abstract: The prediction results of high-performance concrete compressive strength (HPCCS) based on machine learning methods are seriously influenced by input variables and model parameters. This study proposes a method with two stages to select proper variables, simplify parameter settings, and predict HPCCS. The appropriate variables are selected in the first stage by measuring their importance based on random forest, and then are optimized to predict HPCCS in the second stage. The results show that the proposed method was effective for input variable optimization, and could return better predictions than that without variable optimization, provided that the parameters are set within a reasonable range. Compared with previous models, the proposed method shows a strong generalization capacity for HPCCS prediction. We find that the prediction performance of the model is better when the input variables are expressed as absolute mass, and the model performers well when the actual compressive strength of HPC is high.