Study On An Analysis Method For Cluster-based Outlier

Relative to most regular data, the generation mechanisms of outliers are oftendifferent, so outliers may contain important information, and searching of outliers’intensional knowledge has important academic significance and broad application. Theresearch of outliers includes outlier detection and outlier analysis. However, most of theexisting researches of outliers focus on outlier detection, and only few of them analyzeoutliers more. Outliers in the different attribute space will show the differentcharacteristics. In most cases, whether the high-dimensional data is the outlier usuallydepends on the projection of these objects in low-dimensional space. For the formationof outliers, different attributes play different roles. It’s necessary to classify theseattribute subspace, in order to reveal the origin of the outliers.According to this problem, some concepts such as outlier attribute and outliercluster are defined in this paper. Based on the existing outlier mining technologies,some critical theories on outlier dataset were analyzed, such as classifying,characteristic, meaning and origin and an approach to outlier analysis by classifiedoutliers is proposed. Specifically speaking，the main researches of this paper include thefollow aspects:①This paper introduces the theory importance and application value of outlieranalysis in outlier mining, and inspects the domestic and overseas research situation ofoutlier analysis.②The core theories and applicable scopes of outlier mining are analyzed andsummarized roundly, and the current representative methods of outlier analysis aremainly discussed.③Existing classical cluster algorithms are analyzed and compared. Meanwhile thetechniques to detect outliers in clustering algorithm are discussed in detail.④According to the analysis of the relationship between the outliers and clusters, anoutlier analysis method by classified outliers is proposed. The designing mentality andmain content of this method are elaborates and analyzed in detail. Some relatedconcepts are proposed including outlier attribute, trivial outlier, non-trivial outlier andoutlier cluster.⑤Based on the analysis method for cluster-based outlier, an effective cluster-basedoutlier classification algorithm (CBOC) is realized, and experimental results show the effectiveness of the algorithm.⑥Finally, this paper summarizes the main work, analyzes the merits andweaknesses and propose the future work.