[An abnormal characteristic line mining method of celestial spectrum based on attribute weight and w(k)-distance]

Outlier mining is one of the effective methods to find the abnormal celestial spectrum data, and is also one of effective ways to discover the special and unknown celestial bodies. In the present paper, an abnormal characteristic line mining method of celestial spectrum is presented based on the attribute weight and w(k)-distance by using the idea of information entropy. Based on it, an abnormal characteristic line mining system of celestial spectrum was designed and implemented. Firstly, attribute weight of characteristic line was determined by using the idea of information entropy, so that important degree was effectively reflected for each characteristic line. Secondly, massive characteristic line data set of celestial spectrum was reduced by utilizing pruning technique based on neighborhood radius, so that candidate set of abnormal characteristic line was obtained by deleting data objects in which there may not be abnormal characteristic lines. Thirdly, w(k)-distance sum was computed according to the deviation between the data objects in the candidate set, and the objects whose w(k)-distance sum value ranks the first top n were regarded as abnormal characteristic line data objects. In the end, the experimental and the system's running results validated the effectiveness and feasibility of the method by using the SDSS star spectral data set.