DEVELOPMENT OF MODELS FOR PREDICTING CALIFORNIA BEARING RATIO OF LATERITIC SOIL USING SELECTED SOFT COMPUTING TECHNIQUES

Abstract

Models for predicting the California bearing ratio values of lateritic soil was developed using soft computing techniques. Soft computing techniques are algorithm which find provably correct and optimal solutions to problem. The Soaked CBR values used in pavement design takes about 96 hours to complete the test process. This can be time-consuming and expensive, Hence the need for researches to seek for alternate means of obtaining it. Several researchers have employed the use of Artificial Neural network (ANN), Gene expression programming (GEP), Support Vector machine (SVM) and Deep neural network (DNN) to predict CBR values, these models have inherent limitations such as sensitivity to hyper-parameters, limited flexibility and lack of interpretability. This study proposes a new model to address this challenge, Artificial Neural Networks (ANN) and its hybrid (ANFIS) were considered. Soil samples were collected from a burrow pit and required tests were conducted on the collected soil samples, Tests carried out are index, compaction and California bearing ratio. The experimental result data was augmented from data gotten from previous research work (unpublished) in same study area. The result gotten was used for training the models. 70% of the data was used for training and the remaining for the validation of the models. Two different models were developed and the performance of each model was measured by the coefficient of determination (R2), Mean Square Error (MSE) and Root mean square Error (RMSE). Upon analyzing the result, the both models ANN and ANFIS demonstrated high accuracies but ANFIS model gave a higher predictive accuracy of 0.98 as R2, RMSE of 0.11 and MSE of 0.33. ANFIS Model demonstrated exceptional accuracy and precision in capturing complex relationships within the data and hence should be adopted in the prediction of CBR values of lateritic soil.