基于微粒群和子空间的离群数据挖掘算法研究

Outlier Mining Algorithm Based on Particle Swarm Optimization and Subspace

传统的离群数据挖掘方法大多数是利用全局的观点看待离群数据,很难发现低维子空间中的偏移数据。利用微粒群算法(PSO)具有简单、容易实现并且没有许多参数需要调整等优势,提出了一种基于PSO和子空间的离群数据挖掘算法(OM-PSO)。该算法首先将子空间看作微粒,根据偏离数据所在子空间的稀疏系数,采用带有变异算子的PSO算法来搜索子空间,并将子空间中的数据看作为局部偏离数据,即离群数据;最后采用离散化的天体光谱数据作为数据集,实验结果验证了该算法的有效性。

Most methods of traditional outlier mining regard outliers from overall point of view, so it＇s difficulty to find bias data or outliers in subspace. An outlier mining algorithm based on particle swarm optimization and subspace was proposed by using the PSO algorithm＇ characteristics with implementing easily and a few adjustment parameters. The algorithm OM-PSO regards outlier subspace as particle swarm, and searches outlier subspaces with mutational PSO algorithm according to sparsity coefficient of subspace. Finally, the experiment results prove efficient and validity of the OM-PSO algorithm by taking the star spectra data from the LAMOST project.