Bottlenecks and Feasible Solutions of Data Field Clustering in Impact Factor, Time Resolution, Selecting Core Objects and Merging Process

摘要

Data field Clustering is a method to group datasets by virtue of the theory data field which sees every data object as a point with evaluated mass, gets the core data objects, iteratively merges them via simulating the mutual interactions and opposite movements hierarchically. However, there exist some bottlenecks and problems where it may restrict the use and application extending to real areas widely. The determination of impact factor-sigma, the evaluation mass process for every object, the selection of the core objects according to their masses, the ratio of sample initially, time resolution as well as the process of the merging core objects are all crucial to the effectiveness and efficiency of the algorithm results. Through analyzing the main process of data field clustering as well as doing experiment with 2 dimensions data sets, a number of problems are found and several feasible measures to improve the data field clustering is put forward. Using test data sets as example, it is preliminary proven that the improved algorithm obtains a favorable result. Furthermore, the improved method contributes to the further application of data field cluster in Intrusion Detection Systems.