Principal Component Analysis as an Outlier Detection Tool for Polycyclic Aromatic Hydrocarbon Concentrations in Ambient Air

Abstract

Principal component analysis has been used as a tool for the detection of potentially outlying observations in multivariate data sets of polycyclic aromatic hydrocarbon concentrations (PAHs) in ambient air. The outlier statistic developed is the vector distance of each observation at a given site from the origin of principal component space. It is shown that the success of this technique relies on the usually very strong correlation of concentrations of different PAHs in ambient air, such that any deviation from this correlation is noteworthy. Indeed, it is so strong that the first principal component has been omitted from the technique since it is related mostly to absolute concentration. The method has been successful in detecting observations with unusually high concentrations of one or more PAHs. Moreover, it has been possible to identify periods where the UK pollution climate was abnormal during periods of extreme weather. Advice and guidance for the practical use of the technique is also given.