An assessment of adding value of traffic information and other attributes as part of its classifiers in a data mining tool set for predicting surface ozone levels

Abstract

This study seeks to examine to what extent traffic information can improve the prediction of surface ozone levels from mobile sources when coupled with a state of the art air quality monitoring system and the application of data mining tools. For the purpose of the experiment an open-path Deferential Optical Absorption Spectroscopy (DOAS) instrument is used and 10min video samples obtained from Sohar's main highway (SHW) (Sultanate of Oman). This traffic information is collated to recognize, classify, and count three types of vehicles passenger car; light duty vehicle; and heavy duty vehicle. The DOAS is deployed to measure the following gases; ambient nitrogen dioxide (NO2); ozone (O3); sulfur dioxide (SO2); and BTX (benzene, toluene, xylene) across SHW. The ambient concentrations of these gases are measured in situ at time resolutions that vary from 30s to 1min along with simultaneous measurements of meteorological parameters. The Waikato Environment for Knowledge Analysis (WEKA) (Witten and Frank, 2005) software was used for the data mining part of the study. To identify which classifiers in WEKA would be the most suitable in predicting surface O3 levels the following five indexes were used: correlation coefficient (CC); mean absolute error (MAE); root mean square error (RMSE); relative absolute error (RAE); and root relative squared error (RRSE). It was found that the Bagging and M5P classifiers were the most robust when compared to others within the software when measured against the fives indexes. It was identified that with the additions of time and day of the week as well as changing of the parameters as part of the classifiers in WEKA the robustness of the predictions was not enhanced significantly. However, the findings did illustrate that the analysis of traffic information does improve the robustness of the prediction of surface O3 levels.