Research And Application Of Cluster Analysis And Outlier Identification

Outlier identification and cluster analysis are important aspects of data mining, the outliers analysis of data mining algorithms become popular research direction. Most of the outlier analysis algorithm is specific to the operation of static data sets. dynamic data sets can only be taken on the entire data set re-analysis. With the growing volume of data and the increasing demand of data mining algorithms of real-time collection, incremental outlier analysis technology is increasingly of concern.This thesis summarize the data mining, outlier analysis and clustering algorithm and their own major study results, and explaine the detail of the main ideas and the algorithm processes of DBSCAN which is a density-based clustering algorithm and LOF which is a density-based outlier identification algorithm. On this basis, thesis adopted IncrementalLOF which is a local density-based incremental outlier analysis algorithm, and combine with the Network-based Social-security Audit System, experimental results show the LOF and IncrementalLOF in Outlier Analysis of the effect have the consistency, and IncrementalLOF in the large volume of data environment even more superior performance. And IncrementalLOF can be provide into the incremental data mining, draw some unusual, hidding in the data were possible irregularities, such as payment information. It can provide a reliable basis for the social security audit, improve the audit work efficiency, and standardize the business, the reduction of social insurance fraud.