Data Mining Applications In Network Fault Diagnosis

On the basis of studying the key technologies of data mining (DM) and network trouble diagnosing, we put forward the plan to apply data-mining to the diagnosis of network trouble.Data mining can offer many kinds of knowledge, the application of DM in Diagnosing Network Trouble will solve the absence of diagnoseing knowledge. At the same time, some DM methods are both KDD methods and reasoning procedure, for example classification and cluster, so techniques of diagnoses are enriched. The application must bring a new development to Diagnosing Network Trouble.This thesis emphasizes the following work:1. By researching the technology of network trouble diagnoses, we select the object data which will be mined.2. We find the dynamic increment characteristic of object data through analyzing them. Then we apply a method to mining their Association Rules increasingly.3. And meanwhile we recognize that the data is sensitive with time, so we put forward that it is important to add time information in the Association Rules, in this way the availability of the Association Rules has been greatly improved.4. We research the strategy of combining classifiers and adopt such a both serial and parallel structure of combining classifiers.5. We choose three classifiers to construct the multi-classifiers, they are Beyesian network, Support Vector Machine (SVM) and the improved decision-tree: non-absolute decision-tree. We also introduce a simple confusing data method: K nearest neighbor algorithm.6. Use the result of clustering supervising the training of classifier. Introduce an "embedded sample" method, by this method the technology can be directly applied to diagnoses.7. For clustering data, we put forward a new method to computing difference between data which used Universal Gravitation model. At the end, we introduce an algorithm based on gradual generating-tree.