The Cyclone Phase Space (CPS) as an objective method of determining Extratropical Transition (ET) has been used widely. Since the parameters of CPS may not be equal with different resolution reanalysis dataset, they bring errors in deciding the completion time of extratropical transition by using a fixed threshold. In this paper, Radom Forests (RF) method is applied to determine the completion time of TCs over the Western North Pacific from 1979 to 2016, and the factor vector consists of geographical location, center pressure, and CPS parameters. The results of the RF model show that the classification error of RF to distinguish the cyclone into a tropical cyclone or extratropical cyclone is nearly 275% on average, and the percentage of difference between determined and observed completion time within 6 hours is 718% in the training dataset, while the percentage is 659% in the testing dataset, and the percentage of difference within 12 hours exceed 80% in the later dataset. The importance analysis of variables indicates that the greatest contribution to the accuracy of determining is from latitude. The completion time determined by RF lags behind the recorded when the center pressure is lower; however, the determined time is ahead of the recorded as the center pressure is higher and temperature advection at low levels is stronger. In example application, the determined results for Mokakot (2009) and Lionrock (2016) correspond with the completion time recorded well.