Flood Forecasting is a complex procedure because of its uncertainty nature. This problem can be solved by using machine learning techniques. Also short term forecasting with higher accuracy is critical at peak values. Hence, in the current study, using Extreme Gradient Boosting, a rainfall - runoff model has been demonstrated for Kolar Basin, Madhya Pradesh, India. Hourly data of Rainfall and Discharge from 1987-1989 were collected and used for forecasting floods for one to 5 hours ahead. Performance of the model is evaluated using four metrics: Nash Sutcliffe efficiency (NS), Percentage Bias (PBIAS), Root Mean Square Error (RMSE), Coefficient of determination (R2). From the values of the above measures the proposed method has been compared with the Random Forest Model and Support Vector Machine, in which the proposed method shows high accuracy when compared to the existing methods of short-term flood forecasting. The results of this study ensures that Extreme Gradient Boosting method outperforms the Random Forest and Support Vector Machine in terms of Accuracy. © IAEME Publication.