Simulating monthly streamflow using a hybrid feature selection approach integrated with an intelligence model

ABSTRACT

Streamflow prediction is useful for robust water resources engineering and management. This paper introduces a new methodology to generate more effective features for streamflow prediction based on the concept of “interaction effect”. The new features (input variables) are derived from the original features in a process called feature generation. It is necessary to select the most efficient input variables for the modelling process. Two feature selection methods, least absolute shrinkage and selection operator (LASSO) and particle swarm optimization-artificial neural networks (PSO-ANN), are used to select the effective features. Principal components analysis (PCA) is used to reduce the dimensions of selected features. Then, optimized support vector regression (SVR) is used for monthly streamflow prediction at the Karaj River in Iran. The proposed method provided accurate prediction results with a root mean square error (RMSE) of 2.79 m³/s and determination coefficient (R² ) of 0.92.