Machine learning driven optimization of biomedical waste ash concrete for sustainable construction.

Abstract

Concrete produced using ash from biomedical waste is a sustainable construction solution that can help reduce the environmental impact associated with cement production. The study develops a database of Biomedical Waste Ash (BMWA) concrete from literature sources and applies advanced Machine Learning (ML) techniques to predict the mechanical properties of the mixes. Four ML models (Random Forest (RF), Self-Attention and Intersample Attention Transformer (SAINT), TabNet, and an Ensemble model) were subjected to performance evaluation, hyperparameter optimization, and ten-fold cross-validation. RF and TabNet achieved the highest predictive performance, with an R² of 0.82 across strength parameters, while SAINT demonstrated stable generalization but reduced accuracy for certain strength parameters. The ensemble model showed poorer performance than each model, which emphasizes the capability of robust standalone models in specific and limited databases. The external validation showed good agreement between them and hence supports the reliability of our models. Sustainability Index (SI) incorporates cement substitution, durability improvement, and retained strength to evaluate the overall performance of the BMWA concrete at 15% BMWA. The study proposes an integrated data-driven framework that combines advanced ML methods, interpretability analysis, external validation, and sustainability indexing to optimize BMWA concrete and indicate its dual role in reducing the environmental footprint of the cement industry and utilizing biomedical waste in building materials, thus supporting the principles of the circular economy and the global sustainability goals.