<p>Scouring around the bridge foundation removes bed sediment from a riverbed due to obstruction of the pier, threatening infrastructure stability. The scouring phenomenon is further classified as clear water scouring (CWS) and live bed scouring (LBS). CWS occurs when sediment is eroded from the river bed but not replenished by the flowing water. In contrast, LBS occurs when sediment continuously replenishes the scour hole by the approach flow. In this research, previous 648 laboratory and field datasets of CWS and 392 datasets of LBS conditions have been collected to estimate scour depth by utilizing two different machine learning (ML) approaches named multivariate adaptive regression splines (MARS) and extra tree regression (ETR). Five distinct non-dimensional influencing input parameters, such as pier width to flow depth (b/y), pier width to mean sediment size (b/d₅₀), approach mean velocity to sediment incipient velocity (V/V_c), Froude number (F_r) and geometric standard deviation (σ_g) are selected to develop the scour depth prediction model. A Gamma test (GT) has been utilized to identify the optimal combination of input parameters. The findings of present study reported that developed ETR (present model) shows a mean absolute percentage error (MAPE) value of less than 16.0%, a coefficient of determination (R²) value of more than 0.93 for the CWS condition, and for LBS condition value of a MAPE and R² are less than 11.0% and more than 0.88, respectively. Furthermore, it is indicated that the developed ETR (present model) outperforms the MARS (present model) and existing empirical models in CWS and LBS conditions. The significance of the present study is to demonstrate the superiority of these ML models over existing empirical models and identify the most influential CWS and LBS input parameters, which will provide valuable support to bridge engineers in estimating the scour depth.</p>