An online air pollution forecasting system using neural networks

Atakan Kurt, Betul Gulbagci, Ferhat Karaca, Omar Alagha

Research output: Contribution to journalArticlepeer-review

104 Citations (Scopus)

Abstract

In this work, an online air pollution forecasting system for Greater Istanbul Area is developed. The system predicts three air pollution indicator (SO2, PM10 and CO) levels for the next three days (+ 1, + 2, and + 3 days) using neural networks. AirPolTool, a user-friendly website (http://airpol.fatih.edu.tr), publishes + 1, + 2, and + 3 days predictions of air pollutants updated twice a day. Experiments presented in this paper show that quite accurate predictions of air pollutant indicator levels are possible with a simple neural network. It is shown that further optimizations of the model can be achieved using different input parameters and different experimental setups. Firstly, + 1, + 2, and + 3 days' pollution levels are predicted independently using same training data, then + 2 and + 3 days are predicted cumulatively using previously days predicted values. Better prediction results are obtained in the cumulative method. Secondly, the size of training data base used in the model is optimized. The best modeling performance with minimum error rate is achieved using 3-15 past days in the training data set. Finally, the effect of the day of week as an input parameter is investigated. Better forecasts with higher accuracy are observed using the day of week as an input parameter.

Original languageEnglish
Pages (from-to)592-598
Number of pages7
JournalEnvironment International
Volume34
Issue number5
DOIs
Publication statusPublished - Jan 1 2008
Externally publishedYes

Keywords

  • Air pollution
  • Meteorology
  • Neural networks
  • Prediction
  • Turkey

ASJC Scopus subject areas

  • Environmental Science(all)

Fingerprint Dive into the research topics of 'An online air pollution forecasting system using neural networks'. Together they form a unique fingerprint.

Cite this