Incorporating Measurement Error in Astronomical Object Classification

Abstract Most general-purpose classification methods, such as support-vector machine (SVM) and random forest (RF), fail to account for an unusual characteristic of astronomical data: known measurement error uncertainties. In data, this information is often given in the data but discarded because popular learning classifiers cannot incorporate it. We propose a simulation-based approach that incorporates heteroscedastic into existing method better quantify uncertainty classification. The proposed first simulates perturbed realizations from Bayesian posterior predictive distribution Gaussian model. Then, chosen classifier fit each simulation. variation across simulations naturally reflects propagated errors both labeled unlabeled sets. demonstrate use via two numerical studies. thorough simulation study applying procedure SVM RF, which are well-known hard soft classifiers, respectively. second realistic problem identifying high- z (2.9 ≤ 5.1) quasar candidates photometric data. merged catalogs Sloan Digital Sky Survey, Spitzer IRAC Equatorial Spitzer-HETDEX Exploratory Large-Area Survey. reveals out 11,847 identified by without incorporating error, 3146 potential misclassifications with error. Additionally, 1.85 million objects not quasars 936 can be considered new