The rapid integration of green energy resources into the modern grid enhances the power system’s unpredictability and uncertainty, resulting in Low Frequency Oscillation (LFO). Identification of various modes of an oscillating signal is required to maintain stability of power system networks. This study presents an efficient Support Vector Machine-Support Vector Regression (SVM-SVR) based method to recover a corrupted signal measured by phasor measurement units (PMUs) for extracting critical modes from ambient data. The method creates a clustering-based algorithm that predicts the best group of missing data and omits outliers to retrieve clean signal for improved estimation of modes. A MATLAB simulation is carried out to compare results obtained from simulation for the proposed approach with other methods such as Bayesian-K-Medoid and WAR-IQR for synthetic test signals obtained by simulation and real-time probing data of the Western Electricity Coordinating Council (WECC) network to demonstrate the efficacy of the proposed scheme.