基于熵学习机的恒星光谱分类

摘要

数据挖掘被广泛应用于恒星光谱分类.为了提高传统光谱分类方法性能,提出熵学习机(Entropy-based Learning Machine,ELM).在该方法中,熵用来刻画分类的不确定性.为了得到理想的分类结果,分类的不确定性应最小,基于此,可得ELM的最优化问题.ELM在处理二分类问题和稀有光谱发现等方面具有一定优势.SDSS中K型、F型、G型恒星光谱数据集上的比较实验表明:ELM 在进行恒星光谱分类时,其分类性能优于 k近邻(k Nearest Neighbor)和支持向量机(Support Vector Machine)等传统分类方法.%Data mining are widely used in the stellar spectra classification.In order to improve the efficiencies of traditional spectra classification methods,Entropy-based Learning Machine(ELM)was proposed in this pa-per.The entropy was used to describe the uncertainty of classification in ELM.In order to obtain the desired classification efficiencies,the classification uncertainty should be minimized,based on which,we can obtain the optimization problem of ELM.It can be verified that ELM performs well in the binary classification and in the rare spectra mining.Several comparative experiments on the 4 subclasses of K-type spectra,3 subclasses of F-type spectra and 3 subclasses of G-type spectra from Sloan Digital Sky Survey(SDSS)verified that ELM performs better than kNN(k Nearest Neighbor)and SVM(Support Vector Machine)in dealing with the prob-lem of stellar spectra classification on the SDSS datasets.