User-Friendly Clustering for Atmospheric Data Analysis

Atmospheric data analysis is an important area of scientific endeavor, with both government and industrial applications. Our work focuses on clustering particle data acquired via an Aerosol Time-of-Flight Mass Spectrometer (ATOFMS), which is sold and marketed by TSI, Inc. Most papers and software tools developed by the single-particle mass spectrometry community use the ART-2a clustering algorithm. We present in this paper a comparison of the well-known K-means algorithm with ART-2a in this application area. Specifically, we show that despite the entrenched position of the ART-2a algorithm in this domain, Kmeans is faster, more scalable, and considerably easier for practitioners to use while obtaining results of similar accuracy. For data mining practitioners in general and for those who develop software in particular, our work shows that in an important application area K-means is much easier for users to use than ART-2a without sacrificing accuracy. For researchers in the single-particle mass spectrometry community, our experiments demonstrate that ART-2a presents some issues that may be of concern. We propose that K-means offers an attractive alternative.