Influence of Different Samples on Photometric Redshift Estimation for Quasars

Based on the Sloan Digital Sky Survey (SDSS) DR7 and DR12, the UKIRT Infrared Deep Sky Survey (UKIDSS) and the Wide-field Infrared Survey Explorer (WISE), we obtain different wavelength samples and use a kind of tree-based method, extremely randomized trees (Extra-Trees), to estimate the photometric redshifts of quasars. Moreover we compare the performance of this method with k-nearest neighbor algorithm (KNN). Our experimental results show that the accuracy of predicting photometric redshifts is influenced by many factors, such as sample quality, sample selection, feature selection and adopted algorithms. Optimal selection of samples and features contributes to the performance improvement of a regressor. Extra-Trees get better performance than KNN in the low dimensional space while KNN is superior to Extra-Trees in the high dimensional space.