Automatic source detection and classification tools based on machine learning (ML) algorithms are growing in popularity due to their efficiency when dealing with large amounts of data simultaneously and their ability to work in multidimensional parameter spaces. In this work, we present a new, automated method of outlier selection based on support vector machine (SVM) algorithm called one-class SVM (OCSVM), which uses the training data as one class to construct a model of normality in order to recognize novel points. We test the performance of OCSVM algorithm on textit{Wide-field Infrared Survey Explorer (WISE)} data trained on the Sloan Digital Sky Survey (SDSS) sources. Among others, we find $sim 40,000$ sources with abnormal patterns which can be associated with obscured and unobscured active galactic nuclei (AGN) source candidates. We present the preliminary estimation of the clustering properties of these objects and find that the unobscured AGN candidates are preferentially found in less massive dark matter haloes ($M_{DMH}sim10^{12.4}$) than the obscured candidates ($M_{DMH}sim 10^{13.2}$). This result contradicts the unification theory of AGN sources and indicates that the obscured and unobscured phases of AGN activity take place in different evolutionary paths defined by different environments.
Download