Support Vector Machines for Quasar Selection

We introduce an automated method called Support Vector Machines (SVMs) for quasar selection in order to compile an input catalogue for the Large Sky Area Multi-Object Fiber Spectroscopic Telescope (LAMOST) and improve the efficiency of its 4000 fibers. The data are adopted from the Sloan Digital Sky Survey (SDSS) Data Release Seven (DR7) which is the latest world release now. We carefully study the discrimination of quasars from stars by finding the hyperplane in high-dimensional space of colors with different combinations of model parameters in SVMs and give a clear way to find the optimal combination (C-+ = 2, C+- = 2, kernel = RBF, gamma = 3.2). Furthermore, we investigate the performances of SVMs for the sake of predicting the photometric redshifts of quasar candidates and get optimal model parameters of (w = 0.001, C-+ = 1, C+- = 2, kernel = RBF, gamma = 7.5) for SVMs. Finally, the experimental results show that the precision and the recall of SVMs for separating quasars from stars both can be over 95%. Using the optimal model parameters, we estimate the photometric redshifts of 39353 identified quasars, and find that 72.99% of them are consistent with the spectroscopic redshifts within vertical bar Delta z vertical bar < 0.2. This approach is effective and applicable for our problem.