Predicting Concrete Carbonation Depth and investigating the influencing factors through machine learning approaches and optimization

Accurate prediction of the carbonation depth of concrete is very important to protect against harmful consequences such as cracking and corrosion. Nevertheless, due to the complexities of the process and the multitude of available variables, identifying the parameters that are most important in modeling the carbonate depth of concrete is considered a big challenge. This paper deals with the development of a new feature selection method called MOEA/D-ANN. The purpose of this method is to identify the most important variables that help to achieve the highest forecasting accuracy. This proposed method combines the separation-based multi-objective optimization evolutionary algorithm with artificial neural networks to effectively solve the feature selection problem by using the power of optimization methods and machine learning. To evaluate the performance of the introduced method, the algorithm (RReliefF), which is a feature ranking algorithm, has also been used. ANN method has been used to predict concrete carbonate depth and combined MOEA/D-ANN and RReliefF methods have been used to find the influencing variables. The obtained results have shown that the model created using the MOEA/D-ANN approach, by combining the variables determined by it, has a significant reduction in the percentage of errors and an increase in accuracy. In addition, this model reaches the significant value of the coefficient of determination R2 = 0.99, which emphasizes its exceptional accuracy in predicting the depth of concrete carbonate and confirming the accurate selection of influential variables.