A Scientometrics Review of Soil Properties Prediction Using Soft Computing Approaches

In this world, several types of soils are available with their different engineering properties. Determining each soil's engineering properties is difficult because the laboratory procedures are time-consuming. Therefore, several researchers have employed different soft computing techniques to assess soil properties. The soft computing approaches are classified into machine, hybrid, blended, and deep learning. The learning process of these approaches is sub-categorized as supervised, unsupervised, and reinforced learning. This review article presents the performance comparison of different soft computing approaches in predicting the compaction parameters, soaked CBR, unsoaked CBR, and unconfined compressive strength. Several researchers have reported comparisons of the several models and presented optimum performance models based on performance metrics. However, different training databases were utilized in the reported studies. Therefore, the optimum performance model/approach is questionable. In addition, it is well-recognized that the multicollinearity of the training database affects the performance and accuracy of the soft computing models, which has not been studied yet. Very few researchers have performed statistical tests to ensure the quality and quantity of the database. Nowadays, researchers are developing different hybrid approaches to assess soil properties, but the configuration of the hyperparameters is still unknown to obtain the best prediction. This review article introduces several ideas to geotechnical designers/ engineers to develop the optimum performance soft computing model for predicting soil properties. Also, this article will help scientists and researchers get new ideas for innovative research.