This paper presents a system which predicts the grounding topologies of grounding systems (GS) with one, two, three, or four ground rods in straight line, when no prior information about the grounding topology, soil or rod properties is given. This is achieved through the exploration of the information contained into the impulse response of the GS together with the pattern recognition framework. The proposed system is composed by four main elements, an excitation system, a high-speed data acquisition system, a feature extraction, and a pattern recognition model. Experimental setup comprises the evaluation of the accuracy of the proposed system in discriminating among four grounding configurations, composed by the topologies of one, two, three, and four ground rods in straight line. This paper also proposes a new data driven approach to improve the classification performance, in pattern classification models, when dealing with undersampling problem in the acquired signal used to extract the features.