Cluster Analysis of Data from the Particle Analysis by Laser Mass Spectrometry (PALMS) Instrument

Abstract

We describe the use of a hierarchical clustering algorithm on mass spectra of single particles. In this method, the first cluster is found by searching a data set for the two most similar mass spectra. Then the next most similar spectra or clusters are combined sequentially until a stopping condition is met. Chemically reasonable clusters were obtained for several sets of both laboratory and ambient aerosols. Advantages of this technique include assured convergence without multiple iterations, the ability to handle clusters with widely varying numbers of elements, and good performance even when a continuum of points connects different clusters. A disadvantage of this hierarchical clustering is that it operates on a fixed data set and cannot assign spectra to clusters as the spectra are obtained.