California Bearing Ratio estimation using AI: a review of current practices and emerging trends

Abstract

This study investigates the capabilities and reliability of artificial intelligence (AI) tools in estimating the California Bearing Ratio (CBR) for pavement design purposes. The conventional procedure of determining the soaked and unsoaked CBR is time-consuming and cumbersome. Therefore, investigators and scientists developed various AI methods, categorized according to their learning procedure, i.e., machine learning, advanced machine learning, deep learning, and hybrid learning, to compute the soil CBR. This review article compares the performance of the different computational models used in predicting the CBR of soil. It is noted that the performance of these computational models varied due to the use of different databases. Still, the impact of the quality and quantity of the database on predicting the soaked and unsoaked CBR was not analyzed. Additionally, the impact of database multicollinearity on the model’s performance was not analyzed. The study demonstrated that the hybrid learning models are more accurate than the deep and machine learning models in predicting the CBR. Still, configuring the hybrid models (model hyperparameters plus optimization algorithm hyperparameters) is complex and creates structural multicollinearity, which affects the models’ performance and accuracy. The literature showed that the combined effect of structural and database multicollinearity was not analyzed and reported. Therefore, considering these gaps in the literature study, the investigators and scientists can extend the published research and report some interesting outcomes.