Applicability of machine learning algorithms in predicting chloride diffusion in concrete: Modeling, evaluation, and feature analysis

Created by W.Langdon from gp-bibliography.bib Revision:1.8414

@Article{Liu:2024:cscm,

• author = "Weizheng Liu and Guiyong Liu and Xiaolin Zhu",
• title = "Applicability of machine learning algorithms in predicting chloride diffusion in concrete: Modeling, evaluation, and feature analysis",
• journal = "Case Studies in Construction Materials",
• year = "2024",
• volume = "21",
• pages = "e03573",
• keywords = "genetic algorithms, genetic programming, Concrete Durability, Machine Learning Algorithm, Chloride Diffusion Coefficient, Model Evaluation, Feature Analysis, ANN",
• ISSN = "2214-5095",
• URL = "https://www.sciencedirect.com/science/article/pii/S2214509524007241",
• DOI = "doi:10.1016/j.cscm.2024.e03573",
• abstract = "The resistance to chloride diffusion is one of the most crucial durable properties of concrete. However, traditional methods to evaluate this property are time-consuming and inefficient. In this research, backpropagation-artificial neural network (BP-ANN), support vector regression (SVR), genetic programming (GP), extreme gradient boost (XGBoost), and random forest (RF) models were optimised using particle swarm optimisation (PSO) to predict the chloride diffusion coefficient of concretes containing silica fume. A database was also compiled, consisting of various features related to materials composition, curing, and exposure conditions. Statistical assessments were made to evaluate the predictive efficacy of every model. In addition, the distribution of errors and the consistency of each model were scrutinized. The findings indicate that the XGBoost model outperformed the standard models, achieving an R2 value of 0.9382 and an MSE of 3.0162. The models' predictive precision was notably enhanced following their integration with PSO. The PSO algorithm can also decrease the occurrence of significant error points in the predicted values and enhance the consistency of predictive performance across the range of experimental data. Finally, the PSO-XGBoost demonstrated the best comprehensive performance and proved to be the most efficient among the other PSO-synthesised (PSOS) models",

}

Genetic Programming entries for Weizheng Liu Guiyong Liu Xiaolin Zhu

Citations