两阶段无监督顺序前向分形属性规约算法.pdfVIP

计算机研究 与发展 ISSN 1000-1239／CN 11-1777／TP JournalofComputerResearchandDevelopment 45(11)：1955—1964，2008 两阶段无监督顺序前向分形属性规约算法 闫光辉 李战 (西北工业大学计算机学院 西安 710072) (yanguanghui@mail．1zjtu．cn) A TwoPhasesUnsupervisedSequentialForwardFractalDimensionalityReduction Algorithm YanGuanghuiandLiZhanhuai (SchoolofComputerScience，NorthwesternPolytechnicalUniversity，Xian710072) Abstract Boththedimensionality and theamountofdatathatneedstobeprocessedareincreasing rapidlywith the advances in data collection and storage capabilities． Accordingly，reducing the dimensionalityoftheattributevectorstoenhancetheperformanceoftheunderlyingtechniquesisa popularsolution totackletheinfamouscurseofdimensionality．Thefracta1dimensionofonedataset keepsstableastheembeddingdimensionofthedatasetvariesandcanactastheindicatortoguidethe processofthedimensionalityreduction．Therefore，theauthorschoosetheindividualattributefracta1 dimensionandthedifferenceoffractaldimensionaftertheattributemergeoperationasthecriterionof attributecorrelationandtransform thedimensionalityreductionproblem intoanoptimizationproblem whichtriestofindtheattributesubsetwiththemaxima1fractaldimension andtheattributenumber restriction simultaneously．In order to solve theoptimization problem，a two phaseunsupervised sequentialforward fractaldimensionality reduction algorithm is proposed，which integrates the relevanceanalysisprocessandtheredundancyanalysisprocessbasedon thefraetaldimensionofthe individualattributeandtheattributesubset．Theelementarytime—spacecomplexityofthealgorithm is presented．Theexperimentalresultsusingsyntheticandreallifedatasetshow thatthealgorithm gets thesatisfactorysubsetwithratherlOW workloadoffractaldimensioncalculation． Kev words datamining；curseof dimensionality；dimensionality reduction；multifractal；fractal dimension 摘 要 采用单个属性多重分形维数及属性合并之后分形维数变化程度作为属