Predicting the compressive strength of polymer-infused bricks: A machine learning approach with SHAP interpretability

The rapid increase in global waste production, particularly Polymer wastes, poses significant environmental challenges because of its nonbiodegradable nature and harmful effects on both vegetation and aquatic life. To address this issue, innovative construction approaches have emerged, such as repurposing waste Polymers into building materials. This study explores the development of eco-friendly bricks incorporating cement, fly ash, M sand, and polypropylene (PP) fibers derived from waste Polymers. The primary innovation lies in leveraging advanced machine learning techniques, namely, artificial neural networks (ANN), support vector machines (SVM), Random Forest and AdaBoost to predict the compressive strength of these Polymer-infused bricks. The polymer bricks' compressive strength was recorded as the output parameter, with cement, fly ash, M sand, PP waste, and age serving as the input parameters. Machine learning models often function as black boxes, thereby providing limited interpretability; however, our approach addresses this limitation by employing the SHapley Additive exPlanations (SHAP) interpretation method. This enables us to explain the influence of different input variables on the predicted outcomes, thus making the models more transparent and explainable. The performance of each model was evaluated rigorously using various metrics, including Taylor diagrams and accuracy matrices. Among the compared models, the ANN and RF demonstrated superior accuracy which is in close agreement with the experimental results. ANN model achieves R2 values of 0.99674 and 0.99576 in training and testing respectively, whereas RMSE value of 0.0151 (Training) and 0.01915 (Testing). This underscores the reliability of the ANN model in estimating compressive strength. Age, fly ash were found to be the most important variable in predicting the output as determined through SHAP analysis. This study not only highlights the potential of machine learning to enhance the accuracy of predictive models for sustainable construction materials and demonstrates a novel application of SHAP to improve the interpretability of machine learning models in the context of Polymer waste repurposing.