Umudike Journal of Engineering and Technology

Robustness in model selection is pertinent for obtaining parsimonious models, which is highly desirable in the field of science. A Bayesian-based assessment of resilient modulus stress-dependent models used in prediction of resilient modulus as level 2 input in mechanistic empirical pavement design was executed in this study. The assessment was done using the Bayes factor transformation approximation, known as the Bayesian information criterion (BIC), which is a penalised model selection criterion. In order to ensure robustness and possibility of using multiple models, the results of the BIC analysis was interpreted in conjunction with the adjusted coefficient of determination to ensure higher confidence in the model selection process. The study was conducted using data from long-term pavement performance database for two subgrade soils, representing fine and coarse-grained soils. The results obtained indicated the existence of a single robust model for the coarse-grained soil and a confidence set of models for the fine-grained soil. The results of the study clearly indicated the relevance of adopting a robust multimodel selection approach.