This article introduces ARGOS, an automated method for discovering ordinary differential equations (ODEs) from data using sparse regression, denoising, and bootstrap confidence intervals. The method integrates the Savitzky-Golay filter for signal smoothing and derivative approximation, sparse regression for parameter identification, and bootstrap sampling to estimate confidence intervals. ARGOS outperforms SINDy with AIC in identifying ODEs, particularly in three-dimensional systems, by automatically tuning parameters and reducing reliance on manual hyperparameter selection. The algorithm is evaluated on synthetic data from well-known ODEs with varying initial conditions, time series lengths, and signal-to-noise ratios. It demonstrates high accuracy in identifying governing equations, especially under moderate noise levels and sufficient data. ARGOS also performs efficiently in terms of computational time, particularly for larger time series. The method is robust to noise and provides confidence intervals for variable selection, enhancing the reliability of model discovery. The study highlights the importance of data quality and quantity in system identification and shows that ARGOS is more effective than existing methods in handling complex, high-dimensional systems. The approach combines statistical learning with model assessment, offering a reliable framework for discovering governing equations from observational data.This article introduces ARGOS, an automated method for discovering ordinary differential equations (ODEs) from data using sparse regression, denoising, and bootstrap confidence intervals. The method integrates the Savitzky-Golay filter for signal smoothing and derivative approximation, sparse regression for parameter identification, and bootstrap sampling to estimate confidence intervals. ARGOS outperforms SINDy with AIC in identifying ODEs, particularly in three-dimensional systems, by automatically tuning parameters and reducing reliance on manual hyperparameter selection. The algorithm is evaluated on synthetic data from well-known ODEs with varying initial conditions, time series lengths, and signal-to-noise ratios. It demonstrates high accuracy in identifying governing equations, especially under moderate noise levels and sufficient data. ARGOS also performs efficiently in terms of computational time, particularly for larger time series. The method is robust to noise and provides confidence intervals for variable selection, enhancing the reliability of model discovery. The study highlights the importance of data quality and quantity in system identification and shows that ARGOS is more effective than existing methods in handling complex, high-dimensional systems. The approach combines statistical learning with model assessment, offering a reliable framework for discovering governing equations from observational data.