ORIGINAL ARTICLE Usage of output-dependent data scaling in modeling and prediction of air pollution daily concentration values (PM 10 ) in the city of Konya Kemal Polat • S. Savas ¸ Durduran Received: 22 March 2011 / Accepted: 4 June 2011 / Published online: 22 June 2011 Ó Springer-Verlag London Limited 2011 Abstract This paper presents the combination of a data preprocessing called output-dependent data scaling (ODDS) and adaptive network-based fuzzy inference sys- tem (ANFIS) to predict the air pollution daily levels including particulate matter (PM 10 ) concentration values belonging to the city of Konya in Turkey. Also, we have used the regression models including least square regres- sion, partial least square regression, and multivariate linear regression as prediction models in addition to ANFIS model. Data transformation or normalization methods should be used to increase the performance of used pre- diction models and are used prior to prediction algorithms. In this study, we have used the output-dependent data scaling method as data transformation method and com- bined it with ANFIS and regression models. PM 10 con- centration dataset has been taken from Air Quality Statistics database of Turkish Statistical Institute. In PM 10 concentration dataset, the mean values belonging to sea- sons of winter period have been used with the aim of watching the air pollution changes between dates of December, 1, 2003 and December, 30, 2005 in the city of Konya. In the forecasting of PM 10 concentration in Konya province, temperature (°C), humidity (%), pressure (kPa), and wind velocity (km/h) attributes have been used. The experimental results demonstrated that the ODDS method has obtained very promising results in the prediction of PM 10 concentration values. Keywords Output-dependent data scaling  PM 10  Prediction  Adaptive network-based fuzzy inference system  Linear regression models  Air pollution List of symbols MAE Mean absolute error MSE Mean square error IA Index of agreement RMSE Root mean square error P Pressure, kPa R 2 Determination coefficient T Temperature (°C) y m Average of observed points 1 Introduction Particulate matter (PM) is the term used for a mixture of solid particles and liquid droplets hanging in the air. These particles originate from a variety of sources, such as power plants, industrial processes, and diesel trucks, and they are formed in the atmosphere by the transformation of gaseous emissions. Their chemical and physical compositions depend on location, time of year, and weather. Particulate matter consists of both coarse and fine particles [1]. Air pollution causes economic loss and negatively affects the health of human and the life of living beings dependent on increasing objects that are not naturally present in air or objects that amount to no harmful nor- mally. Particulate matter contains very small granular solid K. Polat (&) Electrical and Electronics Engineering Department, Abant I ˙ zzet Baysal University, Go ¨lkey Campus, 14280 Bolu, Turkey e-mail: kemal_polat2003@yahoo.com S. S. Durduran Department of Geomatics Engineering, Faculty of Engineering and Architecture, Selcuk University, 42035 Konya, Turkey e-mail: durduran@selcuk.edu.tr 123 Neural Comput & Applic (2012) 21:2153–2162 DOI 10.1007/s00521-011-0661-z