Predicting Depth of Concrete Carbonation through Machine Learning and Optimization: Investigation of Influential Factors.

Accurate prediction of the depth of concrete carbonation is paramount in safeguarding againstdetrimental outcomes like cracking and corrosion. Nevertheless, due to the intricacies of the process andthe multitude of variables at play, discerning the critical parameters that hold the utmost significance inmodeling concrete carbonation depth poses formidable challenges. This paper presents the development ofa novel feature selection method called MOEA/D-ANN. The objective of this method is to identify the mostcrucial variables that contribute to achieving the highest prediction accuracy. The proposed approachcombines the MOEA/D optimization algorithm with Artificial Neural Networks (ANN) to effectively solvethe feature selection problem by leveraging the power of both optimization and machine learningtechniques. To evaluate the performance of the introduced technique, a conventional feature selectionmethod, Regression Relief Feature Selection (RReliefF), was also employed. The ANN method wasemployed to predict the depth of concrete carbonation, and separate models were developed using theinfluential variables identified by the MOEA/D-ANN and RReliefF methods. The findings reveal that themodel created using the MOEA/D-ANN approach, incorporating variables determined by it, demonstratessignificantly reduced errors. Moreover, the model attains an impressive R2 value of 0.99, underscoring itsexceptional precision in forecasting the depth of concrete carbonation and validating the meticulousselection of influential variables.

Predicting Depth of Concrete Carbonation through Machine Learning and Optimization: Investigation of Influential Factors. Keywords:

concrete , carbonation depth , machine learning , optimization