RUI: Guiding Gamma-Ray Burst Classification with the KDD Process

点击翻译按钮获取中文摘要

AST 0098499HakkilaDespite tremendous scientific breakthroughs during the past two decades, gamma-ray burst (GRB) astro-physics is still in its infancy. GRBs are the most energetic events in the universe. Further understanding of the GRB phenomenon rests in part on understand-ing distinct behaviors that can be associated with physical mechanisms. Identification of distinct physical behaviors (e.g. classification) is an important component of the scientific method. However, distinct behaviors do not always indicate the presence of separate source populations. The complex, overlapping properties of GRBs are a case in point: they have long confounded efforts to subclassify their behaviors. Many identified behaviors have been shown to result from either instrumental or sampling biases. GRB classification can be better carried out using statistically and computationally rigorous approaches of Knowledge Discovery in Databases (KDD) combined with a detailed understanding of instrumental and sampling biases. Recently, there have been claims that certain attributes correlate with burst lumi-nosity. The claims, however, are based on a very limited data set of GRBs with afterglows, and on attributes that have only been defined for a small subset of bursts observed by BATSE (the Burst And Transient Source Experiment on NASA's defunct Compton Gamma-Ray Observatory). Are the correlation's between luminosity and these attributes self-consistent with observations of the large BATSE data set? Do instrumental biases play any role in these apparent correlations? All GRBs with afterglows thus far belong to the long class of GRBs; is there similar evidence for these behaviors in the short class of GRBs? Do these attributes indicate the presence of other GRB subclasses? These questions will be addressed in this project using a large GRB database, a set of well-defined and appropriate attributes, detailed knowledge of the instrument(s) from which the observations are made, and KDD methodology. After this database has been produced, pattern recognition algorithms will be applied to GRB classifica-tion. Based on preliminary results, it is expected that general GRB subclasses will be identified, as well as substructures indicative of specific GRB behaviors.As a part of this project, a database of complex preprocessed GRB attributes will be developed and made available via the World Wide Web. Funding for this project was provided by the NSF program for Extragalactic Astronomy & Cosmology (AST/EXC).***