Abstarct: Nowadays, transmission line protection has become increasingly important due to the expansion of the power network. Current differential protection is one of the important protections of transmission lines, which is used in critical lines such as compensated lines, GW transmission lines, and lines prone to high impedance faults. This protection is baxsed on data transfer between differential relays at both ends of the line via communication channels. This issue makes this protection vulnerable to cyber-attacks. Fast and accurate detection of cyber-attacks is essential because line protection plays a critical role in maintaining the network's stability and reliability. In this thesis, a new method baxsed on the cumulative ensemble bagging tree classifier is proposed to improve the accuracy and speed of detecting cyber-attacks from internal faults in the line current differential protection. In this method, the features of the ellipse fitted on the Lissajous curve formed by the currents at the beginning and end of the protected line are used. These features include the major and minor axes of the ellipse, its rotation angle, and the deviation of its center from the coordinate origin, which are extracted after calculating the general coefficients of the ellipse. To evaluate the performance of the proposed method, IEEE 9-bus and IEEE 39-bus test systems are used. The simulation results show that the proposed method can detect all types of internal faults with low and high impedances, up to a resistance of 5000 Ohms, from false data injection attacks and time synchronization attacks under various conditions, including normal conditions, load switching, and capacitor bank switching. Additionally, the proposed method have proper performance in noisy conditions and changes in sampling frequency. The simulation results in both networks demonstrate that the proposed method can distinguish internal faults from cyber-attacks under various conditions, achieving an accuracy of 99.95% in an average time of 7.53 ms in the IEEE 9-bus test system, and an accuracy of 99.91% in an average time of 10.45 ms in the IEEE 39-bus test system.