Automated Classification of Transient and Variable Sources
Abstract
Better and faster technology is increasing the amount of collected data in many scientific fields, all of them requiring a common task: extract knowledge from massive, multi parametric data sets, as rapidly and efficiently possible. Most systems today rely on a delayed human judgment and this “manual” approach will simply not scale to the next generation of surveys thus the need of a machine learning approach. We present a sleeping expert framework that makes use of many different ensembles of classifiers (eg, kNN, Neural Networks, Bayesian Networks, Decision Trees) that can be activated according to the input data; results from the single classifiers are then combined in order to achieve a better classification rate. Using data sets extracted from the ongoing Catalina Real-Time Transient Surveys (CRTS), we show the results obtained applying this framework to different astronomical problems (eg, classifying transients on the basis of features describing the light curves, systematic search of CV). Moreover, we illustrate a variety of feature selection strategies used to identify the subsets that give the most information; given the high number of parameters this is quickly becoming a crucial task in analyzing astronomical data sets.
- Publication:
-
American Astronomical Society Meeting Abstracts #221
- Pub Date:
- January 2013
- Bibcode:
- 2013AAS...22135220D