AI-Assisted Optimization of Coconut-Pith–Reinforced Fly-Ash Geopolymer Composites: Integrating Experimental Design, Machine Learning, and Bayesian Optimization for Sustainable Binder Development

• A sustainable NaOH-activated geopolymer composite was developed using coal fly ash and coconut pith fibers. • The composite exhibited enhanced strength and compactness through optimized NaOH concentration and fiber content. • Advanced characterization (FTIR, XRD, XRF) confirmed the formation of stable aluminosilicate and reinforcing mineral phases. • Machine learning with Bayesian optimization predicted optimal mix parameters, streamlining material design and minimizing resource use. The rapid growth in construction demand has intensified the need for sustainable binders that reduce the environmental burden associated with Portland cement. This study develops a sodium hydroxide (NaOH)-activated coal fly-ash geopolymer composite reinforced with coconut pith, an abundant agricultural by-product, to enhance mechanical performance and material sustainability. A combined experimental and machine learning (ML) framework was implemented to predict and optimize the unconfined compressive strength (UCS) of the composite. Fifteen experimental formulations were used to train ten regression models representing diverse learning biases, and an ensemble surrogate model was subsequently coupled with Bayesian optimization (BO) to identify optimal mix parameters. The BO-predicted formulation (10.8 M NaOH, 3.2 % pith) corresponded closely with the experimentally measured optimum (10 M NaOH, 3 % pith, UCS = 18.32 MPa), confirming the model’s predictive validity. The results demonstrate that integrating ML and BO enables efficient exploration of the compositional space, reducing experimental effort while improving data-driven material design. This hybrid approach establishes a scalable, resource-efficient pathway for developing low-carbon geopolymer binders that valorize agricultural waste and advance sustainable construction technologies.