Improving Embankment Settlement Predictions using Artificial Intelligence: Applications in Geotechnical Engineering

In road geotechnics, predicting embankment settlement is one of the most critical challenges, including management of completion in time, forecasting costs, and optimizing technical solutions at the design stage, especially when related to compressible soil types. Nonetheless, the nature of soil behaviour is such that heritage methods used for calculating and predicting soil settlement and deformation under load are often unreliable and tend to have only a limited capability with respect to the estimation of actual settlement. In this paper, a machine learning-based approach is introduced to enhance the prediction of deformation in structures on compressible soil. The proposed technique involves the application of an Artificial Neural Network (ANN) and density-based clustering and ordering (DBSCAN) to a real database collected during construction monitoring for high-speed train works in Morocco. DBSCAN was very effective for the organization and processing of databases, as well as a useful tool for identifying predominant spatial patterns of occupation and erroneous measurements caused by the heterogeneities and complexities found in compressible soils.