Prediction of Fracture Initiation in Cohesive Soil Using a Data-Mining Approach

The determination of fracturing pressure and failure mode in cohesive soils is a controversial issue that seems to be no consensus among previous studies. In this paper, a novel data-mining approach based on the XGBoost algorithm for predicting fracturing initiation in cohesive soils was proposed. A machine learning framework, including a regressor and a classifier that both have 14 input features and one output variable, was developed for fracturing pressure prediction and failure mode classification, respectively. The engineering verification indicates that the proposed approach paves a new way to predict fracturing pressure and failure mode in cohesive soils, which could provide some guidelines for engineering application.