The high impedance arc fault (HIAF) poses a significant threat to the living being as it involves arcing. The enormous amount of heat generation during arc is a major concern in this regard. There are different types of arc that may occur depending on the arcing conditions and involved surfaces. The severity of the arc is determined by the involved arcing surface. In this study, arc in metallic (sphere gap, rod gap) and nonmetallic (leaning tree, concrete) surfaces on a distribution system is considered for the analysis. An empirical mode decomposition (EMD)-based approach is applied along with k-nearest neighbor (KNN) for the characterization and classification of the real-time arc of different arcing conditions. The results obtained using EMD and KNN algorithm on arc signals successfully characterize and classify different HIAF by their harmonic signature. Along with KNN, the cross-validation data-mining algorithm is also applied to check the robustness of the approach. © 2019 John Wiley & Sons, Ltd.