Machine Learning-Based Regression Modeling of Compressive Strength Behavior in RESIN8 Recycled Concrete Containing Organic Contamination

This study investigated the compressive-strength behavior of concrete containing RESIN8 recycled plastic, paying particular attention to food waste contamination. A total of 36 specimens representing conventional concrete, concrete containing uncontaminated RESIN8, and concrete containing RESIN8 with 5% organic contamination were tested after 14 and 35 days of curing. The control concrete exhibited the highest mean compressive strength at both curing durations. Among the RESIN8-containing mixtures, the contaminated mixture showed modestly higher mean strengths than the uncontaminated mixture; however, this observation was limited to the specific experimental conditions examined. Two-way analysis of variance (ANOVA) indicated significant effects of mixture group and curing duration, as well as their interaction. Five regression models, namely linear regression (LR), multilayer perceptron (MLP), radial basis function network (RBFN), random forest (RF), and support vector regression (SVR), were evaluated using six-fold leave-one-set-out (LOSO) cross-validation. The models achieved R2 values of 0.739–0.825 using seven explanatory variables and 0.850–0.874 using four composition- and curing-related variables. Among the models evaluated, SVR yielded the highest observed cross-validated R2 under the present conditions. The LR-based analysis assigned the highest relative importance values to curing duration and the sand-to-RESIN8 replacement effect. Correlation and variance inflation factor (VIF) analyses confirmed perfect collinearity between sand and RESIN8; consequently, their independent effects could not be separated. These results indicate the potential applicability of machine learning (ML)-based regression modeling within the investigated RESIN8 concrete mixtures, although the predictive findings should be interpreted within the limited experimental domain of this study.