Chemically-assigned classification of aerosol mass spectra

ClusterSculptor: Software for expert-steered classification of single particle mass spectra

“…Other clustering approaches classify particles into classes on the basis of a set distance metric [1,3,7]. In all cases, at the end of the classification process each class is represented by an average/representative mass spectrum.…”

“…Past experience with different clustering approaches [1,3,4,9,10,12] reveals that classification of single particle mass spectra could be greatly improved if the users aid the clustering process by utilizing their scientific knowledge to steer the classification.…”

“…Other approaches rely on using different clustering algorithms [10] or classification parameters [9,10]. Some have defined discriminant chemical markers or characteristic peaks [1,9] that are used in particle assignment. In other cases data are even modified prior to classification [4,9,10].…”

Mass Spectrometric Methods for Arosol Composition Masurements

“…An alternative approach used by some groups is to analyse spectra sequentially using pattern matching algorithms such as the ART-2A (Adaptive Resonance Theory) neural network (Bhave et al, 2001;Carpenter et al, 1991;Phares et al, 2001;Song et al, 1999) and ADAMS (Algorithm for Discriminant Analysis of Mass Spectra) (Tan et al, 2002). These can group particle spectra according to similarities with previously seen spectra and can adapt to unexpected results and outliers by creating new groups on the fly or progressively altering the reference spectra.…”

The EDAM project: Mining atmospheric aerosol datasets

“…There are, on the other hand, approaches that allow one to incorporate domain knowledge into clustering algorithms. 96,128,132,133 Such techniques would allow users to "guide" clustering algorithms toward categories of interest. Finally, variations of scalable clustering algorithms might lend themselves to on-line clustering in order to provide information on classes found while the experiment is running.…”