Investigating the Mechanical and Microstructural Properties With Machine Learning Optimization of Recycled Eco‐Friendly PET Modified Concrete

Over the past decades, studies have demonstrated the effectiveness of polyethylene (PE) and PE terephthalate (PET) as an alternative to aggregates in enhancing concrete performance. This research aims to predict the compressive strength (CS) of PE and PET‐incorporated concrete by evaluating the influence of key factors, including cement content, PET and PE dosage, supplementary cementitious materials (SCMs), water content, aggregate composition, superplasticizer (SP), and curing age. A dataset consisting of 300 mix proportions and their corresponding strengths was analyzed using advanced machine learning (ML) techniques, such as extreme gradient boosting (XGB), categorial boosting (CatBoost), gradient boosting (GB), decision tree (DT), bagging regressor (BAG), and random forest (RF). The data were partitioned into training and testing sets, followed by statistical evaluations to assess the correlations between input parameters and CS. Furthermore, to strengthen the findings of this study, specimens were prepared using various dosages of PET as a sand replacement (10%, 20%, 30%, and 50%) for experimental evaluation. Among the models tested, XGB and CatB exhibited the highest predictive accuracy, achieving R 2 values of 0.99 for the training data and 0.92 for the testing data. The SHapley Additive exPlanations (SHAP) and partial dependence plots (PDPs) indicated that water content and curing age had a significant impact on CS. These findings underscore the reliability of XGB and CatB models as robust predictive tools for assessing the CS of PET‐incorporated concrete, contributing to the advancement of sustainable construction practices. Furthermore, microstructural evaluation reveals the inner structure of PET concrete and provides insights into the distribution of PET particles within the matrix, which directly influences its CS.