CPT Data Interpretation Employing Different Machine Learning Techniques

The classification of soils into categories with a similar range properties is fundamental geotechnical engineering procedure. At present, this based on various types cost- and time-intensive laboratory and/or in situ tests. These soil investigations are essential for each individual construction site have to be performed prior the design project. Since Machine Learning could play key role reducing costs time needed suitable investigation program, basic ability models classify from Cone Penetration Tests (CPT) evaluated. To find an appropriate model, 24 different models, three algorithms, built trained dataset consisting 1339 CPT. applied algorithms Support Vector Machine, Artificial Neural Network Random Forest. As input features, combinations direct cone penetration test data (tip resistance qc, sleeve friction fs, ratio Rf, depth d), combined “defined”, thus, not directly measured (total vertical stresses σv, effective σ’v hydrostatic pore pressure u0), used. Standard classes grain size distributions behavior according Robertson as targets. compared respect their prediction performance required learning time. best results all targets were obtained using Forest classifier. For distribution, accuracy about 75%, Robertson, 97–99%, was reached.