Scientific Research and Essays

  • Abbreviation: Sci. Res. Essays
  • Language: English
  • ISSN: 1992-2248
  • DOI: 10.5897/SRE
  • Start Year: 2006
  • Published Articles: 2755

Full Length Research Paper

Ranking water quality variables using feature selection algorithms to improve generalization capability of artificial neural networks

Ayhan Samandar
Department of Civil, Düzce Vocational School, Düzce University, 81620, Düzce, Turkey.
Email: [email protected]

  •  Accepted: 19 April 2010
  •  Published: 04 June 2010

Abstract

Artificial neural networks (ANNs) have been recently used intensively in the water resource management and planning. As the demand on the clean water increases, the estimation of important parameters using ANNs to evaluate the surface water quality has become an importance practice. In this study, feature selection methods, which are important tools in data mining, are used to shed light into a further understanding in the correlation between measured parameters of water quality variables and dissolved oxygen level. The aim is to enhance the generalization ability of ANNs in prediction of dissolved oxygen level. Three different feature selection methods are utilized to discern the inherent correlation in the water quality variables in a given data set collected from Melen River, Turkey. Results show that, a small set of variable can be identified, which effectively improves the generalization ability of ANNs.

 

Key words: Water quality, artificial neural networks, Melen river, ranking.