Galaxy and quasar classification based on local mean-based k-nearest neighbor method

With the implementation of some large sky surveys such as LAMOST, We will collect massive spectra data. The automated spectral classification becomes more and more important. In this paper, a local mean based k-nearest neighbor (LMKNN) method is used for the automated classification of galaxies and QSOs (quasars). The main idea of LMKNN is that it selects k nearest neighbors from each class of training samples firstly, and then computes the mean vectors of the k nearest neighbors in each class. Finally, classifies a sample into the class with the distance of the mean vector to the sample. In the experiment, KNN, LMKNN and SVM were compared with the real galaxy and QSO spectra data from the LAMOST-DR1. The experimental results show that the method of LMKNN performs better and at least as well as than KNN, SVM in terms of the correct classification rates. It can achieve a better classification rate as high as 98.97%, 97.58% for galaxies and QSOs respectively. The average correct classification rate can also achieve 98.33%.