CPT-Based Soil Classification through Machine Learning Techniques: Fid-Bau Portal

CPT-Based Soil Classification through Machine Learning Techniques

Faraz Athar, Mohammad / Khoshnevisan, Sara / Sadik, Laith

Soil classification is a fundamental step prior to any design procedure in every geotechnical project. However, the soil testing required for classification purposes is expensive, time consuming, and resource-intensive; making it difficult to conduct accurate and enough testing on a routine basis. Engineers and researchers have carried out soil testing for many years, leading to the development of a huge dataset. Artificial Intelligence (AI) and Machine Learning (ML) which have become more popular in the civil and geotechnical engineering fields in the past few years provide a great opportunity to use the collected data for the development of evolutionary prediction models. In this paper, machine learning algorithms are used to build a model which can classify the soil using Cone Penetration Testing (CPT) data, focusing on three regions in the US (southeastern, central, and western). Random Forest, Support Vector Machine, K Nearest Neighbors, and Extreme Gradient Boosting algorithms are the four ML approaches used in this study for the prediction of the Soil Behavior Type (SBT) zone classification which is widely used for classifying the soil behavior based on CPT results. The performance of the adopted models is compared using different accuracy metrics and the confusion matrix. In this study, the Random Forest classifier and Extreme Gradient Boosting models are found to perform better in terms of soil prediction and total model training.