Comparison of clustering methods for identification of outdoor measurements in pollution monitoring

Abstract

This paper considers the problem of post-processing air pollution data to clearly identify outdoor clusters, by removing indoor data and "noise" caused by air from indoors mingling with air from outdoors. In this paper, several different clustering algorithms are compared using data from measurements in Macao. It is shown that X-means generally outperforms the others for this purpose and can successfully separate data modified by noise. Such a technique simplifies the collection of large data sets since the person taking the measurements does not have to make any advance decisions about what is pure outdoor, or pure indoor, data. However, it is also shown in this work that setting up suitable procedures can be quite complex.