Automatised classification of WISE sources: first results, future prospects

TL;DR

This paper demonstrates an automated method using Support Vector Machines to classify galaxies, stars, and quasars in WISE mid-infrared data, achieving high accuracy with promising results for future large-scale surveys.

Contribution

It introduces a novel SVM-based classification pipeline tailored for WISE data, optimizing parameter selection and demonstrating high classification performance.

Findings

Achieved up to 95% completeness and purity for quasars.

Classifier performance depends on magnitude, color, and aperture magnitude.

Promising results with only three classification parameters.

Abstract

We present the first results of our dedicated programme of automatised classification of galaxies, stars and quasars in the mid-infrared all-sky data from the WISE survey. We employ the Support Vector Machines (SVM) algorithm, which defines a hyperplane separating different classes of sources in a multidimensional space of arbitrarily chosen parameters. This approach consists of four general steps: 1) selection of the training sample, 2) selection of the optimal parameter space, 3) training of the classifier, 4) application to target data. Here, as the training set, we use sources from a cross-correlation of the WISE catalogue with the SDSS spectroscopic sample. The performance of the SVM classifier was tested as a function of size of the training set, dimension of the parameter space, WISE apparent magnitude and Galactic extinction. We find that our classifier provides promising results already for three classification parameters: magnitude, colour and differential aperture magnitude. Completeness and purity levels as high as 95% are obtained for quasars, while for galaxies and stars they vary between 80-95% depending on the magnitude, deteriorating for fainter sources.