Optimizing shear strength characterization of engineered emulsion-stabilized cold recycled materials for pavement design through predictive statistical modelling

Fundamental shear strength parameters, cohesion and frictional angle, offer significant insights into the shear characteristics of cold recycled materials and are good indices for mechanistic-empirical design models of cold recycled pavements. Material preparation, testing procedures and analysis are, however, often labor-intensive, costly, and time-consuming, necessitating efficient and practical alternatives for parameter estimation. This study employs a modeling framework to predict the shear strength characteristics of engineered emulsion-stabilized cold recycled materials. Data from monotonic triaxial shear tests were modeled using two machine learning approaches, namely, (1) self-validating ensemble modelling, which integrates multiple linear regression (forward selection, adaptive least absolute shrinkage and selection operator, elastic net) & neural network configurations, and (2) support vector regression. Based on model performance evaluations through a range of statistical error metrics, support vector regression demonstrated high accuracy and generalizability, identifying mixture coarseness and density as primary factors influencing cohesion, while stabilizer content and chemical additive presence were important for frictional angle. The predicted parameters were further validated within the deviator stress ratio framework, showing strong alignment with experimentally computed results. As an outcome of this study, a user-friendly Microsoft Excel-based predictive tool was developed to facilitate efficient shear parameter prediction and support pavement design and evaluation.