Predictive modeling assessment of advanced concrete materials incorporating plastic waste as sand replacement

Abstract This research aims to contribute to the advancement of sustainable construction materials using a new composite of coated plastic waste as sand replacement material. This research assessed the predictive capabilities of Random Forest (RF), Particle Swarm Optimization-Support Vector Regression (PSO-SVR), and a Genetic Algorithm Optimized Artificial Neural Network (GA-ANN) that enable accurate, data-efficient prediction of compressive strength in plastic-waste foamed concrete, reducing experimental overhead and guiding sustainable mix optimization to forecast the compressive strength of foam concrete containing plastic waste. The models were evaluated using R 2 metrics, where RF scored 0.9872 and 0.9005, and GA-ANN scored 0.9979 and 0.8853 for the training and testing sets, respectively. Sensitivity analyses of the RF and GA-ANN models were conducted to evaluate the compressive strength of the foam concrete and the impact of each associated input parameter. The findings confirmed that both models accurately predicted the compressive strength of the material. The R 2 values for both models were calculated: for RF 0.9872 and 0.9005, and for GA-ANN 0.9979 and 0.8853. Sensitivity analysis indicated that the highest Permutation Importance values for cement, foam, sand, water-to-cement ratio, and plastic waste were 0.39, 0.34, 0.17, 0.11, and 0.39, respectively. In the GA-ANN case, the greatest Permutation Importance Values of 0.41, 0.31, 0.13, 0.11, and 0.05 were assigned to cement, sand, water-to-cement ratio, foam, and plastic waste, respectively, in that order concerning compressive strength. The PSO-SVR model in green maintained a good balance (AUC = 0.97 in training and AUC = 0.93) in testing. The PSO-SVR model achieved an average performance between those of the other two models. The MAE value was approximately 1.5 in training and 2.8 in testing, whereas the RMSE value was in the range of 4.5–5.0. The results showed the practicality of AI-based frameworks in the focus optimization of mix design and multi-criteria prediction of performance metrics of sustainable foam concrete containing recycled plastic waste.