An efficient machine learning approach for predicting concrete chloride resistance using a comprehensive dataset

Maedeh Hosseinzadeh; Seyed Sina Mousavi; Alireza Hosseinzadeh; Mehdi Dehestani

doi:10.1038/s41598-023-42270-3

An efficient machine learning approach for predicting concrete chloride resistance using a comprehensive dataset

Sci Rep. 2023 Sep 12;13(1):15024. doi: 10.1038/s41598-023-42270-3.

Authors

Maedeh Hosseinzadeh¹, Seyed Sina Mousavi¹, Alireza Hosseinzadeh¹, Mehdi Dehestani²

Affiliations

¹ Faculty of Civil Engineering, Babol Noshirvani University of Technology, 484, Babol, 47148-71167, Iran.
² Faculty of Civil Engineering, Babol Noshirvani University of Technology, 484, Babol, 47148-71167, Iran. dehestani@nit.ac.ir.

Abstract

By conducting an analysis of chloride migration in concrete, it is possible to enhance the durability of concrete structures and mitigate the risk of corrosion. In addition, the utilization of machine learning techniques that can effectively forecast the chloride migration coefficient of concrete shows potential as a financially viable and less complex substitute for labour-intensive experimental evaluations. The existing models for predicting chloride resistance encounter two primary challenges: the constraints imposed by a limited dataset and the absence of certain input variables. These factors collectively contribute to a decrease in the overall effectiveness of these models. Therefore, this study aims to propose an advanced approach for dataset cleaning, utilizing a comprehensive experimental dataset comprising 1073 pre-existing experimental outcomes. The proposed model for predicting the chloride diffusion coefficient incorporates various input variables, such as water content, cement content, slag content, fly ash content, silica fume content, fine aggregate content, coarse aggregate content, superplasticizer content, fresh density, compressive strength, age of compressive strength test, and age of migration test. The utilization of the artificial neural network (ANN) technique is also employed for the processing of missing data. The current supervised learning incorporates both regression and classification tasks. The efficacy of the proposed models for accurately predicting the chloride diffusion coefficient has been effectively validated. The findings indicate that the XGBoost and SVM algorithms exhibit superior performance compared to other regression prediction algorithms, as evidenced by their high R2 scores of 0.94 and 0.91, respectively. In relation to classification algorithms, the findings demonstrate that the Random Forest, LightGBM, and XGBoost models exhibit the highest levels of accuracy, specifically 0.93, 0.96, and 0.97, respectively. Furthermore, a website has been developed that is capable of predicting the chloride migration coefficient and chloride penetration resistance of concrete.