Evaluation of input variable selection methods in artificial neural networks for estimating missing daily precipitation

Abstract

This study evaluates different Input Variable Selection (IVS) methods for precise daily precipitation estimation using artificial neural network (ANN) models. The effectiveness of the models is measured using the average Nash-Sutcliffe efficiency (NSE) from 50 runs per model. Initially, a Multiple Linear Regression and a base model with all sixteen parameters resulted in test NSE scores of 0.390 and 0.656, respectively. Subsequently, the Self-Organizing Map (SOM), Analytic Hierarchy Process, K-Nearest Neighbour, and K-Means methods were used to select input variables of different types. The SOM method achieved its best performance with a test NSE of 0.692 when four variables were selected, while the other methods reached their peak performance with the same NSE of 0.718 when two similar variables were selected. This highlights the importance of IVS methods in using ANN models to overcome overfitting and achieve an efficient model for precise precipitation estimation.

Keywords:

• artificial neural network
• missing data
• precipitation
• input variable selection
• self-organizing map
• analytic hierarchy process

Disclaimer

As a service to authors and researchers we are providing this version of an accepted manuscript (AM). Copyediting, typesetting, and review of the resulting proofs will be undertaken on this manuscript before final publication of the Version of Record (VoR). During production and pre-press, errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal relate to these versions also.