Hybrid machine learning approach for prediction and design optimization of marshall stability in graphene...

Marshall Stability (MS) is a key, yet costly and time-consuming, metric for designing asphalt concrete (AC) in general, and Graphene Oxide (GO)-modified AC in particular. To address this, this study introduces a novel hybrid machine learning framework by coupling a classical Gradient Boosting (CGB) model with a Sailfish Optimizer (SFO) to accurately predict and optimize the MS of GO-modified AC. A comprehensive database of 33 experimental outcomes was curated to develop and validate the model, with performance assessed using statistical benchmarks like the correlation coefficient (R). The proposed CGB model optimized by SFO demonstrated exceptional predictive prowess, yielding an impressive R-value of 0.975 on the unseen test data. A subsequent sensitivity analysis revealed that aggregate gradation (specifically sieve percentages for 2.36, 9.5, and 4.75 mm) exerts the most pronounced influence on MS. The primary contribution of this work is a validated computational tool that allows engineers to rapidly and reliably predict the performance of novel GO-modified asphalt, significantly accelerating design optimization and reducing laboratory costs.