Piers in water environments are susceptible to local scour, particularly when there are two piers arranged eccentrically. The flow of water causes erosion of the sediment surrounding the piers, resulting in distinct flow patterns near the eccentric pier. Researchers are now using data-driven soft computing algorithms to more correctly compute local scour at isolated piers. However, there is limited study on reliable formulas for determining maximum scour depth (sdm), and even little experimental data exists for eccentric arrangements of piers. For this study, 50 data sets are collected from previous experimental studies conducted during the last decade. To determine the sdm in clear water, ensemble frameworks such as the Light Gradient Boosting Machine (LGBM) and Random Forest Regression (RFR) are used. The seven parameters that are considered independent variables include flow shallowness, flow intensity, sediment gradation, sediment coarseness, time, flow skew angle, and spacing between piers. Three indicators of performance are used to evaluate machine learning models (MLMs) such as Coefficient of Determination (CD), Mean Absolute Error (MAE), and Mean Squared Error (MSE). Two equations from the literature were compared to the current MLM based on the stated performance indicators. The study’s findings show that LGBM outperforms RFR in both training and testing. In addition, it is used LGBM and RFR to assess the relevance of features. The Taylor skill score compared the newly developed model to the literature model, demonstrating that LGBM is superior to computing the scour depth.