基于GEP的流数据分类压缩并行算法研究

Parallel Classification Compression Algorithm for Stream-Data Based on Granular Analysis and Storage of GEP

针对数据挖掘中流数据分类精度不高,数据压缩率低的问题,提出一种基于粒度分析与转存式GEP的流数据分类压缩并行算法,实现流数据的快速并行分类压缩。首先使用粒度分析技术对流数据的属性求得极小粒度集,根据划分规则得到近似粒度空间,然后根据不同类型流数据建立不同的G EP分类器模型,最后采用动态转存记录集形式,将数据送至GEP压缩模型实施压缩。再将串行算法扩展到M PI+OpenM P混合编程模型后得到其并行算法,采用UCI数据、通讯账单验证算法的性能。实验结果表明,分类压缩效果耗时较理想,压缩比效果明显,其中在校学生通讯账单耗时在96 s左右,压缩比达到1/3。

Considering the low accuracy of the stream-data classification hasn＇t high accuracy and com-pression rate for data mining,the stream-data parallel classification compression algorithm was proposed based on granular analysis and storage of GEP in order to achieve faster parallel classification compres-sion algorithm of streaming data. Firstly, get the least set of stream-data with the granular analysis method ,and the approximate granular space according to division rules. Secondly, establish correspond- ing GEP classification model for different stream-data^Finally,send the data to compression model of GEP and compression data with dynamic storage record set form^extend serial algorithm to the parallel algorithm in MPI＋OpenMP hybrid programming model,and verify the algorithm performance with the UCI data and communications bill. The experimental result shows that the effect of the classification compressions time-consuming and the compression ratio are satisfactory, the student＇s communication bill time-consuming is about 96 s,and the compression ratio can be achieved to 1/3.