In the field of Geophysics and civil engineering applications, Ground Penetrating Radar (GPR) technology have become one of the emerging Non-Destructive Testing (NDT) methods for its ability to perform tests without damaging structures. In this context, applications like concrete rebars assessments, utility networks surveys or precise localisation of embedded cylindrical pipes still remain challenging. Inversion of geometrical parameters such as depth and radius of embedded cylindrical pipes and dielectric parameters of its surrounding material are of great importance for preventive measurements or quality control. Innovative signal processing techniques associated with GPR could perform physical and geometrical characterisation. In this paper, Performances of supervised machine learning method on GPR data are evaluated. Support Vector Machines (SVM) classifications, Support Vector Machines regression (SVR) and Ray-based methods are used in order to correlate information about radius and depth of the embedded pipes with the velocity of stratified media, in various numerical configurations. The approach is based on the hyperbola trace emerging in a set of B-scans, whereas the shape of hyperbola highly varies with depth and radius of the pipe as well as the velocity of the medium. In the ray-based method, inversion of the wave velocity and the pipe’s radius was performed through an appropriate nonlinear least mean squares inversion technique. The features selection within machine learning models also performed on the information picked from observed hyperbola’s travel times. Simulated data are obtained by the Finite-Difference Time-Domain (FDTD) method with the numerical tool 2D GprMax. The study is carried out at mono-static, ground coupled configuration. A parametric comparison is also included in the analysis of performances of the mentioned techniques in terms of relative errors estimation against designed parameters.