Plastic waste in concrete: Data-driven insights into strength behavior

Rapid urbanization and global population growth lead to a rise in plastic waste (PW) worldwide, making recycling a global challenge. A sustainable approach using PW in concrete industry may minimize the adverse effects. Integration of PW in concrete impacts mechanical performance, particularly compressive strength (CS). In addition, identifying the exact contribution of plastic remains difficult, particularly due to its heterogeneous characteristic and complex interactions within the concrete matrix. Hence, this study provides a comprehensive machine learning (ML) technique to predict the CS of plastic aggregate concrete (PAC). Considering this, a wide range of datasets comprising 612 experimental test outcomes was compiled from 74 literature sources. Seventeen widely adopted ML models including linear, tree-based, ensemble, and neural network algorithms were developed using eight influential input parameters. Among all models, MLPNN outperformed showing highest accuracy (R 2 = 0.826), with more than 75% of training and 70% of testing predictions falling within a minor error range. The narrow 95% Confidence Interval of ±0.9928 MPa further demonstrates low prediction variability. The sensitivity analyses using SHAP, ICE, and feature importance reported that dry unit weight, W/C ratio, and plastic aggregate (PA) addition are most impactful parameters. Results show that PA replacement up to 5% can maintain or slightly enhance CS, whereas higher additions (up to 20%) may reduce strength by around 5 MPa. This investigation offers a sustainable approach to concrete design using PW with the help of ML, which can be further enhanced through a GUI-based practical application and utilized by designers for PAC. • Seventeen ML models used to predict compressive strength of PAC. • Dataset of 612 experimental results compiled from global studies. • MLPNN achieved best results, followed by ET, GB, CB, and SVR models. • Dry Unit Weight, W/C ratio, and PA addition were key influencing factors for prediction. • SHAP analysis confirmed Dry Unit Weight as the most dominant feature.