Prediction of unconfined compressive strength of cement–lime stabilized soil using artificial neural network

The present research aims to enhance the stability of soil by utilizing varying dosages of cement and lime. Additionally, the study seeks to create a predictive model based on Artificial Neural Network to estimate the unconfined compressive strength (UCS). During the investigation, the materials under examination underwent essential engineering tests, including microstructural characterization and the UCS test. The results from the UCS test indicated a consistent increase in strength values as the curing time and cement content were raised. To develop the predictive model for UCS, ANN-based models with sigmoid function with different architectures were constructed. A multiple regression model was also used for comparison. The training dataset comprised 80 data points, while the testing dataset contained 20 data points. The data set was divided into 3 parts: training, testing and validation. From data set, 60% data were used for training, and 10 and 30% were used for validation and testing, respectively. The outcomes of the study demonstrated that the ANN model 4 (8-16-32-1) utilizing the feed forward Levenberg–Marquardt (trainlm) backpropagation function outperformed all other models, achieving an R value of 0.89 during training and 0.7 during testing. In summary, this research focuses on stabilizing soil by employing cement and lime, while also developing an effective ANN-based model to predict the unconfined compressive strength of soil. The study showcases the superiority of the ANN model with the feedforward Levenberg–Marquardt (trainlm) backpropagation function and underscores the significant influence of cement content on the UCS prediction. © 2023, The Author(s), under exclusive licence to Springer Nature Switzerland AG.