Machine learning formulation for predicting concrete carbonation depth: A sustainability analysis and optimal mixture design

Predicting the Carbonation Depth (CD) of concrete is crucial for evaluating its long-term durability and resistance to environmental degradation, particularly in structures exposed to Carbon Dioxide (CO 2 ). Such phenomenon leads to corrosion of embedded reinforcement and structural deterioration . In this study, Multi-Gene Genetic Programming (MGGP) and Random Forest (RF), as Machine Learning (ML) techniques, were used to predict the CD of various concrete samples containing fly ash, an eco-friendly partial substitute for cement, using a dataset of 198 concrete mixtures. The dataset consists of the content of cement, fine and coarse aggregates, fly ash, water, superplasticizer , and their respective testing conditions, including relative humidity , CO 2 concentration, and the cycle duration of the carbonation test. However, identifying the most relevant variables to enhance prediction accuracy is a challenging and complex task. Therefore, to tackle this challenge, the Pareto Envelope-based Selection Algorithm II (PESA-II) was integrated with Artificial Neural Networks (ANNs) to form a multi-objective optimization approach. This novel feature selection method efficiently identifies the most important variables, leading to improved predictive accuracy. Although recent studies have used various ML algorithms to predict carbonation depth, none of them have generated a predictive formula for such variable. In this regard, the MGGP method was selected due to its efficiency to generate such formula. This formula demonstrated high accuracy, achieving an R-squared value of 0.91 on the testing set and 0.92 on the training set, indicating the capability of this method. Finally, the Grey Relational Analysis (GRA) method was employed as a Multi-Criteria Decision-Making (MCDM) tool to identify the optimal concrete mixture, considering both cost efficiency and environmental impact, including CO 2 emissions and energy consumption associated with the production of each concrete ingredient.