In hydrology, accurate predictions and monitoring of river discharge are critical for river engineering, flood mitigation, water resource management and agricultural purposes. The Brahmaputra-Jamuna in Bangladesh, one of the highest discharge rivers in South Asia, is fundamental to the region's socio-economic structure and a major driver of flooding. Hence, predicting discharge in this river is crucial for managing water resources, protecting infrastructures and minimizing flood risks. In this study, four machine learning models, namely, Random Forest (RF), Support Vector Machine (SVM), Gradient Boosting Regression Tree (GBRT), eXtreme Gradient Boosting (XGB) and their ensemble model were employed for weekly river discharge (Qt) prediction at Bahadurabad Transit. Weekly observed discharge and ERA5 reanalysis rainfall data with lag times from 1976 to 2022 were used for model calibration and validation. Various graphical and statistical evaluation metrics were employed to assess the model's performance. The findings indicate Rt, Rt-1, Qt-1 is the most effective inputs in predicting discharge (r = 0.92). Individual and ensemble models have a very good performance (R2, NSE: 0.85 to 0.92), and the ensemble model outperforms RF by 4.55%, SVM by 8.24%, GBRT by 3.37% and XGB by 2.22%. For peak discharge simulation, the ensemble model shows the best performance (R2 = 0.94, NSE = 0.94, RMSE = 4013.11 m3/s and MAE = 2843.60 m3/s). The reliability analysis verified the ensemble's superiority. The models in this study were efficient, adaptable and applicable for river discharge prediction in the Brahmaputra-Jamuna and other river gauge stations.