Analyzing the viability of agro-waste for sustainable concrete: Expression-based formulation and validation of predictive models for strength performance

Abstract

By lowering environmental impact and increasing resource efficiency, agro-waste in concrete provides a sustainable substitute for traditional materials. This study focuses on optimizing concrete mix designs using agricultural byproducts such as corn cob ash, palm oil fuel ash, rice husk ash, sugarcane bagasse ash, and wheat straw ash, alongside water-cement ratio and curing duration. An experimental dataset was compiled, targeting compressive strength (CS) and tensile strength (TS) as output parameters. Two advanced machine learning (ML) techniques, gene expression programming (GEP) and multi expression programming (MEP), were applied for predictive modeling and optimization. Model accuracy was assessed using statistical metrics and Taylor’s plots, while SHapley Additive exPlanations analysis was used to interpret input parameter influence. Results showed that MEP outperformed GEP in both CS and TS predictions. For CS, the MEP model achieved an R2 of 0.963 compared to 0.934 for the GEP model, while for TS, the MEP model reached 0.961 compared to 0.945 for GEP. The study highlights the role of ML in enhancing mix design efficiency, reducing trial-and-error experimentation, and accelerating the development of sustainable, high-performance construction materials. Such studies demonstrate how ML can streamline concrete mix design by minimizing experimental trials, saving both time and resources. By incorporating agro-waste materials, they provide engineers with practical tools to develop sustainable, high-performance concretes that lower environmental impact while maintaining structural reliability.