Abstract
Alkali-activated concrete (AAC) has emerged as a viable sustainable alternative for the building of roads and rigid pavements, particularly in India, where infrastructure development is a major priority. With the exponential growth of highway networks, there is a noteworthy emphasis on enhancing mechanical performance of concrete pavements to overcome their inherent brittleness and limited load-carrying capacity. This research examines the incorporation of nano-silica (NS) and nano-alumina (NA) to improve the mechanical properties of pavement quality alkali-activated concrete (PQAC). Additionally, polyvinyl alcohol fibres (PVAF) and Polypropylene Fibre (PPF) were integrated into the concrete mix to address the brittle nature of PQAC and improve the tensile strength of concrete. Given the challenges associated with optimising these material combinations, this research also leverages advanced machine learning models, including Multilinear Regression (MLR), Decision Tree (DT), Random Forest (RF), AdaBoost, Support Vector Regression (SVR), Gradient Boosting (GB), to predict the split tensile strength (STS) of PQAC. A thorough analysis of predicted performance was carried out utilising assessment measures. The findings demonstrate that the AdaBoost model outperforms other models in terms of test performance, achieving an R2 value of 0.79. This surpasses the R2 values of MLR (0.61), SVR (0.6), DT (0.79), GB (0.791) and RF (0.796). The remaining four error measures have the lowest values among all models, with MSE = 0.202, RMSE = 0.45, CV=0.714 and MAE = 0.38. The study highlights the superior performance of ensemble models in accurately predicting STS, underscoring their potential as reliable tools for optimising material compositions in pavement applications and thereby supporting or partly replacing laboratory split tension tests, thereby saving time and cost. This research contributes to the broader goal of developing more durable and sustainable concrete mixes for construction projects.