River meandering is a complex process involving the complex of bank erosion, sediment transport, and flow through channel bends. In order to expand the capacity of the conveyance system and for better flood management, the flow is divided between the main channel and the floodplains. When a river overflows its bank during flood, part of load is carried by bordering floodplains. The relationship between the flow in the main channel and the floodplains has a significant impact on the discharge in a compound meandering channel. The accuracy of flood prediction is essential for giving the flood warning, calculating the flood danger and managing the rivers over the long term. This paper attempts to develop a model to calculate the discharge prediction in the meandering compound channel using machine learning techniques such as Extreme gradient boosting (XGBOOST) and Categorical boosting (CATBoost) by considering the impact of several geometric, flow, and roughness parameters. XGBoost and CATBoost are powerful gradient boosting algorithms known for their ability to handle structured data and categorical variables. Results show that the both XGBoost and CATBoost predicted the discharge (Qp) satisfactorily with the coefficient of determination (R2) value higher than 0.90 and mean absolute percentage error (MAPE) below 10% for training and testing datasets. However, XGBoost model prediction accuracy is better compared to the CATBoost model.