Research On Network Traffic Collection Model Based On Data Mining

With the rapid develoment of Internet, the incessant increment of network flows and the quick growth of application services, which make network topology and user behaviors more and more complicated, network traffic collector is faced with the challenge. It has been the important researchs in network field to know performance of network, analyze network behavior and defense virus invading and hakers attacking. The original packets are needed by this researchs. So network traffic collector plays a more and more important role in network research. Network traffic collection technology is analyzed and explored in this thesis and a Data Mining-based Multi-Agent Distributed Network Traffic Collection Model is presented. The Agent system of DA_DNTC consists of three main modules: transport service agent, data mining agent, packet capture agent. Every agent works independentlty and together to accomplish traffic collection. The introduction of the data mining technique gives the study ability to the traffic collection system, this ability makes the system to study and extract eigenvalue from the network, then enlarge DA_DNTC's rule database, the ability to traffic collection can be enforced ceaselessly. With several characteristics such as self-adaptive, distribute and application oriented, DA_DNTC system can solve many problems in current traffic collection system.A bright prospect can be seen of DA_DNTC which is merged by data mining and agent technology. Applying data mining technique in traffic collection can reduce workload of manual analysis. However data mining technique also has some problems, such as effiency and accuracy. This thesis proposed data mining algorithm to enhance the effiency of the system. Clustering algorithm of data mining is analyzed detailedly. According to the characteristic of enormous network data set, an algorithm for clustering which has higher efficiency is proposed and is proved feasible through the experimentation. At last, research is done on the association rule mining algorithm. The most efficient algorithm is chosen from the two popular association rule mining algorithms—Aporiori and FP-Growth by the comparison and analysis. The FP-Grown algorithm outperforms the Apriori algorithms that used in the DA_DNTC system.