COMPARISON OF DATA SPLITTING METHODS FOR NEURAL NETWORKS USED IN HYDROLOGICAL FORECASTING

In past decades, artificial neural networks (ANN) have been applied frequently in the prediction of water resources parameters. While developing ANN models, the available data were split into training, validation, and testing subsets. Most researchers did not consider the impact of the splitting method on the performance of the models and split the data using ad hoc methods. To investigate the impact of the data splitting method used on the performance of ANN models for inflow prediction, four data splitting methods were investigated to develop an ANN that predicts the inflow for multiple reservoirs. These methods were compared regarding the bias and variance of the performance of the models. The results indicate that the performance of the model is indeed influenced by the choice of the data division method. It was found that self-organizing maps (SOM) and k-means based stratified sampling are well suited for data division in the ANN development for water inflow prediction.

Journal

FRESENIUS ENVIRONMENTAL BULLETIN

Year

2022

Volume

31

Issue

1

Pages

34-42

ISSN

1018-4619

eISSN

1610-2304

WoS UID

WOS:000736347700004