Predicting California bearing ratio using machine learning to model soil behavior for road construction in Tshimoyapula, Botswana

Abstract

The present research work is carried out to predict California Bearing Ratio (CBR) values of soils from the Tshimoyapula area near Serowe in the Central District of Botswana based on the index properties using machine learning techniques. The CBR test is a very important and common test performed on soils to assess the stiffness modulus and shear strength of subgrade materials to determine the thickness of overlaying layers in pavement design. It is an expensive and timeconsuming test in addition to difficulty in keeping the sample in the desired condition. The construction industry is one of the least digitized in the world, and using Artificial Intelligence could help achieve profitability, efficiency, safety, and security. Machine learning techniques, namely, Regression Analysis (RA) and Artificial Neural Network (ANN) were developed with different configurations using various laboratory soil properties comprising Liquid Limit (LL), Plastic Limit (PL), Plastic Index (PI), Maximum Dry Density (MDD), and Optimum Moisture Content (OMC) of 200 soil samples that laboratory CBR test was performed on. The index properties were used as input parameters for different models with the CBR as output. Results indicate a good correlation between the input parameters and the output. Artificial Neural Networks showed the least error and the highest accuracy followed by Linear Regression among the models.