Ground Penetrating Radar systems have been successfully used to access concrete structures conditions. Moreover, inclusions in concrete can be discriminated by simple models based on traces obtained by GPR. In this work, concrete blocks with different inclusions were probed in controlled conditions. Some features were extracted from Ascans of this experimental data set. To get efficient models, raw data were submitted to features selection and space reduction methods. Without complex data pre-processing, good accuracy and more explainable models with less computational burden were obtained.