Flexural performance evaluation using computational tools for plastic-derived mortar modified with blends of industrial waste powders

Abstract Cementitious composites, which include waste plastic, have limited uses in the construction sector due to a decline in strength. Substituting industrial waste for cement in plastic-based mortar mixes (PBMs) is an environmentally preferable and long-term sustainable alternative. Machine learning (ML) models including bagging regressors (BGR), AdaBoost regressors (ADB), and random forest regressors (RFR) were used to assess the flexural strength (FS) of PBMs. An extensive dataset of 408 samples was employed that included six inputs such as cement, fine aggregate, plastic, silica fume, marble powder, and glass powder. Analysis was performed to determine the importance of contributing variables using the Shapley additive explanations (SHAP) approach. When it came to predicting PBM’s FS, RFR outperformed BGR and ADB. The RFR model had an R 2 -value of 0.98, which was much higher than 0.88 and 0.92 reported by the BGR and ADB models, respectively. The statistical results indicated that BGR had an MAPE of 4.00%, an ADB of 2.90%, and an RFR of 1.80%, adding further weight to the claims made about the deployed ML models. Sand, cement, and silica fume were identified as the three most critical components in terms of strength according to the SHAP study. The created models can assist in identifying the feasible solutions of PBMs for diverse input parameter values, hence minimizing the necessity for expensive and time-intensive laboratory testing.