In this research a novel communication based high impedance fault (HIF) protection scheme is proposed on the basis of wavelet packet transform (WPT) and extreme learning machine (ELM). The scheme is applied to the protection of transmission and distribution lines considering evolving and cross-country faults. Fault detection and clearing are the most important tasks of all protection systems, however, simultaneous and complex faults are the cases that most of the conventional protection systems are incapable of making appropriate decisions. The bottlenecks of the conventional protection systems are more highlighted when the faults are HIF. In this study, WPT, which is one of the useful methods in signal processing, is utilized to extract high frequency coefficients of three phase currents of both ends of transmission line. Furthermore, ELM which is an artificial neural network based classifier, is employed to detect the faulted phases. Consequently, the proposed protection algorithm will be compared with the distance protection as a conventional protection system and also with a recently proposed HIF protection algorithm to verify its robustness, security and dependability. The proposed method is not sensitive to the location, inception angle and impedance of the faults, power swing, power system topology and noise.