The forecasting of runoff or watershed modeling mainly depends on consistency and accuracy of observed data, climate variability, and precise availability of parameters, which are critical aspects of rainfall-runoff simulation for semi-arid regions. The problem will increase in the coming future due to climatic changes and frequency of occurrence. Recently, a data-driven machine learning (ML) approach that relies on data to build models and to make predictions is being used effectively for various applications. In this study, five ML models, artificial neural network (ANN), long short-term memory (LSTM), support vector machine (SVM), CatBoost, and Gaussian process regression (GPR) are used in predicting monthly stream flows at three-gauge stations in Godavari River basin. Statistical methods such as coefficient of correlation (R), coefficient of determination (R2), and Nash Sutcliffe efficiency (NSE) are used to check the accuracy of the models. Among the five models, the LSTM model better predicts the monthly streamflow at all gauge stations. R2 value for the LSTM model at all the gauge stations is between 0.75 and 0.92. The results suggest that LSTM is a promising tool for future runoff prediction studies in Godavari River basin.