Enhancing stone matrix asphalt performance with sugarcane bagasse ash: Mechanical properties and machine learning-based predictions using XGBoost and random forest

This study investigates the incorporation of Sugarcane Bagasse Ash (SCBA) as an alternative filler in Stone Matrix Asphalt (SMA) mixtures to enhance their mechanical and durability characteristics while promoting environmental sustainability. SCBA, a by-product of the sugar industry, was added at varying proportions (0%, 2%, 4%, 6%, 8%, and 10%) to partially replace conventional mineral filler. A series of laboratory tests—including Marshall Stability and Flow, Indirect Tensile Strength (ITS), and Moisture Susceptibility (TSR)—were conducted to evaluate the performance of the SCBA-modified SMA mixtures. The results revealed that the inclusion of 6% SCBA yielded the most favorable outcomes. Marshall Stability increased significantly (up to 9.4 kN), ITS improved to 943 kPa, and moisture susceptibility was enhanced, demonstrating a higher tensile strength ratio compared to the control mixture. These improvements are attributed to the high silica content and pozzolanic nature of SCBA, which contributed to better bonding and matrix densification. Importantly, the modified mixtures remained within acceptable flow and volumetric criteria, indicating practical viability. In parallel, the study applied machine learning (ML) models—Support Vector Regression (SVR), Random Forest (RF), and Extreme Gradient Boosting (XGBoost)—to predict the mechanical properties of SMA based on input mix parameters. Among the models, XGBoost achieved the best performance (R² > 0.93), followed closely by RF, while SVR showed relatively lower accuracy. The ML predictions closely matched experimental outcomes, demonstrating the potential of data-driven modeling in optimizing asphalt mix design. This dual approach—experimental validation and ML-based prediction—confirms that SCBA can effectively serve as a sustainable filler in SMA, offering performance gains while reducing environmental impact. Additionally, the successful integration of ensemble ML algorithms provides a reliable framework for forecasting key performance metrics in asphalt engineering. The findings support broader adoption of industrial by-products in pavement materials and the use of artificial intelligence for efficient material design.