Daily Inflow and Outflow Prediction for Kabini Dam using Random Forest Regression

This work presents a machine learning-based predictive model for estimating the daily inflow and outflow of the Kabini Dam, employing a Random Forest regression technique. The model uses historical hydrometerological and time-based features, including dam water level, rainfall, and date-derived components. Data from 2020 to 2023 were collected from the Karnataka Water resource department and preprocessed through missing value imputation, feature extraction, and target cleaning. Two models-one each for inflow and outflow, were trained using an 80/20 split and evaluated using Mean Squared Error(MSE), Root Mean Squared Error(RMSE), Mean Absolute Error(MAE), and Coefficient of Determination(R2). The inflow model achieved MSE = 82.3 (cumecs)2, RMSE = 9.07 cumecs, MAE = 6.21 cumecs and R2 = 0.93, while the outflow model reported MSE = 95.8 (cumecs)2, RMSE = 9.79 cumecs, MAE = 6.84 cumecs and R2 = 0.91. Compared to a baseline persistence model, Random forest provided a reduction in RMSE. The models were deployed through a Flask API, enabling real-time predictions for decision support. The results establish Random Forest as a robust, operationally feasible method for dam flow forecasting and management.