基于模糊C均值聚类的天文光谱特征线软离散化

连续数值属性离散化是天文光谱数据预处理中的主要研究内容之一。针对天文光谱特征线,提出了一种基于改进模糊C均值聚类的天文光谱特征线软离散化算法。该算法首先利用样本的密度值选取特征线的候选初始模糊聚类中心,有效地克服了对噪声数据敏感的缺陷;其次采用决策表中的相容性作为评判标准,动态的调节聚类参数,以达到优化的光谱特征线离散化效果;最后采用晚型星、类星体、高红移类星体SDSS天文光谱特征线数据集。实验验证了该算法具有较高的识别率,为天文光谱特征线数据预处理提供了一种新途径。

A soft discretization method of celestial spectrum characteristic line based on fuzzy C-means clustering

Discretization of continuous numerical attribute is one of the important research works in the preprocessing of celestial spectrum data. For characteristic line of celestial spectrum, a soft discretization algorithm is presented by using improved fuzzy C-means clustering. Firstly, candidate fuzzy clustering centers of characteristic line are chosen by using density values of sample data, so that its anti-noise ability is improved. Secondly, parameters in the fuzzy clustering are dynamically adjusted by taking compatibility of decision table as criteria, so that optimal discretization effect of the characteristic line is achieved. In the end, experimental results effectively validate that the algorithm has higher correct recognition rate of the algorithm by using three SDSS celestial spectrum data sets of high-redshift quasars, late-type star and quasars.