Input variable selection for water resources systems using a modified minimum redundancy maximum relevance (mMRMR) algorithm

Mohamad I. Hejazi, Ximing Cai

Research output: Contribution to journalArticle

Abstract

Input variable selection (IVS) is a necessary step in modeling water resources systems. Neglecting this step may lead to unnecessary model complexity and reduced model accuracy. In this paper, we apply the minimum redundancy maximum relevance (MRMR) algorithm to identifying the most relevant set of inputs in modeling a water resources system. We further introduce two modified versions of the MRMR algorithm (α-MRMR and β-MRMR), where α and β are correction factors that are found to increase and decrease as a power-law function, respectively, with the progress of the input selection algorithms and the increase of the number of selected input variables. We apply the proposed algorithms to 22 reservoirs in California to predict daily releases based on a set from a 121 potential input variables. Results indicate that the two proposed algorithms are good measures of model inputs as reflected in enhanced model performance. The α-MRMR and β-MRMR values exhibit strong negative correlation to model performance as depicted in lower root-mean-square-error (RMSE) values.

Original languageEnglish (US)
Pages (from-to)582-593
Number of pages12
JournalAdvances in Water Resources
Volume32
Issue number4
DOIs
StatePublished - Apr 1 2009

Keywords

  • Input selection
  • MRMR
  • Modeling
  • Mutual information
  • mMRMR

ASJC Scopus subject areas

  • Water Science and Technology

Fingerprint Dive into the research topics of 'Input variable selection for water resources systems using a modified minimum redundancy maximum relevance (mMRMR) algorithm'. Together they form a unique fingerprint.

  • Cite this