A combination index measurement in forecasting daily air pollution index

Abstract

Error magnitude is a measurement commonly used in forecast evaluation. However, the purpose of forecasting air quality is to maintain the air quality within assigned guidelines. Thus, the index measurement is important to be considered. But, the problem arises when the index is used to gauge the values of different offices and these measurements are found to be degenerate in commonly occurring situations. Therefore, this study aims to overcome both of the limitations. The daily air pollutant index (API) data from year 2005 to 2011 was used to compare the forecast performance between Box-Jenkins methods, artificial neural networks (ANN) and hybrid method. The forecast accuracy measurements used include mean absolute percentage error (MAPE), root mean squared error (RMSE), mean absolute deviation (MAD), true predicted rate (TPR), false positive rate (FPR), false alarm rate (FAR) and successful index (SI) including the proposed index measurement, combination index (CI). It is found that the index measurement enhance the ability to measure the air quality forecast performance in choosing the best forecast method with CI significantly overcome the limitation of existing index measurement. Thus, this study suggests to use the appropriate measurement in accordance to the purpose of forecasting.