The Gandak River is a crucial lifeline for agriculture and the livelihoods of local communities in Bihar, India. However, its fluctuating water levels, particularly during intense rainfall, pose a severe risk of flooding. This flooding can lead to crop damage, displacement of communities, and significant economic losses. Understanding and accurately predicting the river’s water levels is essential for effective flood management, ensuring timely interventions, and safeguarding the region’s agricultural productivity and public safety. The present study focused on predicting water levels in the Gandak River Basin using random forest (RF), support vector regression (SVR), and extreme gradient boosting (XGBoost) models. The models were trained and tested using 20 years of data (2004 to 2023) for short- and medium-term prediction horizons. It is observed that random forest moderately outperforms XGBoost in one-day-ahead predictions; however, overall, XGBoost outperforms both SVR and RF for five-day and ten-day lead time forecasts. The results suggest that XGBoost is a reliable model for both short- and medium-term water level prediction, with R2 values ranging from 0.977 (for 1 day) to 0.925 (for 10 days).