In recent years, sub-/super-synchronous oscillation frequently occurred because of the rapid development of inverter-based resources and high-voltage direct current transmission. Online monitoring and identifying sub-/super-synchronous oscillation are thus of great importance for the safe and stable operation of power systems. For this reason, a method of sub-/super-synchronous oscillation mode identification based on the amplitude-frequency characteristics of synchronous phasor data is proposed. First, the influence mechanism of that oscillation on the data of a phasor measurement unit (PMU) is analyzed. The results show that the positive and negative spectrum of PMU data is linearly correlated with the mode of the oscillation. Then, the coherent spectrum of multi-point PMU data is used to distinguish between oscillation and noise. This effectively reduces the misjudgment caused by noise. Further spectral analysis is conducted on PMU data under sub-/super-synchronous oscillation. Four amplitude-frequency characteristic values are established, and the feature set of oscillation data is used as input to train and optimize the eXtreme Gradient Boosting (XGBoost) model. This model establishes the mapping relationship between amplitude-frequency features and oscillation patterns. The proposed method uses the inherent amplitude-frequency characteristics of PMU data in oscillation environments, along with the powerful generalization and computational efficiency of the XGBoost algorithm. As a result, it achieves rapid and accurate identification of sub-/super-synchronous oscillation patterns in noisy environments. Both simulations and field tests demonstrate the effectiveness and usefulness of the proposed method.