Using neural networks to predict peak Cryptosporidium concentrations

Gail Montgomery Brion, T. R. Neelakantan, Srinivasa Lingireddy

Research output: Contribution to journalArticlepeer-review

32 Scopus citations

Abstract

Neural network modeling was used to examine the relationships between multiple interrelated water quality and quantity parameters at the intake to a water treatment facility located on the Delaware River. The relationships were used to train a neural network model to predict peak concentrations of Cryptosporidium oocysts at the intake of a New Jersey water treatment facility. Input parameters to the model were selected based on their correlation with oocyst concentrations and stepwise evaluation of neural network training. The final trained neural network model predicted two conditions of input Cryptosporidium concentrations - background and above background (assigned as 1 and 0, respectively) - from eight other water quality parameters. Clostridium perfringens concentrations were the most significant input parameter in predicting the final model's performance. Turbidity was the least significant parameter. Furthermore, a site-specific, linear relationship between the numbers of full oocysts and the total number of oocysts recovered by the Information Collection Rule method at this water treatment plant intake was noted (full oocysts = 0.595×total oocysts, R2 = 0.9011).

Original languageEnglish
Pages (from-to)99-105
Number of pages7
JournalJournal / American Water Works Association
Volume93
Issue number1
DOIs
StatePublished - Jan 2001

ASJC Scopus subject areas

  • General Chemistry
  • Water Science and Technology

Fingerprint

Dive into the research topics of 'Using neural networks to predict peak Cryptosporidium concentrations'. Together they form a unique fingerprint.

Cite this