This paper presents a signal processing and machine learning-based approach to classify different types of arcs due to the interaction of a medium voltage distribution line and different surfaces. Different kind of arcing surfaces, i.e., concrete, wet-sand, grass, and leaning tree, are considered in a real-time environment to create different arcs. The similarity found in various arcing events is the low (in mA) current flowing during the arc. The voltage signals are taken as the basis of the whole analysis. The signal processing technique used in this study is empirical mode decomposition (EMD). The results obtained by the application of EMD along with different support vector machine (SVM) techniques on voltage signals successfully classifies various high impedance arc faults (HIAFs) for various arcing surfaces based on their harmonic footprints. © 2021, Springer Nature Singapore Pte Ltd.