Research On Application Of Association Rule Mining To Telecommunication Network Alarm Analysis

With the scale of communication networks becomes increasingly large and the structure becomes increasingly complex, they can produce a large number of alarms every day. Thus, the traditionally human treatment of alarms cannot meet the demands of enterprises. Data mining is a new technique that can be used to extract the hidden information from oceans of data in large databases or data warehouses, so it has become an effective tool to help isolate and diagnose network faults, choose correct measures and assume the task of pre-maintenance and trend analysis.This paper is aimed to study the relevant data mining algorithms based on the characteristics of alarm data. That involves the study of the literatures and comparative analysis of various algorithms. As a result, association rule mining algorithm is selected as the subject of the research. As for the fact that the existing algorithms are always inefficient and produce too many rules, this paper is aimed to propose an improved temporal association rule mining algorithm. Firstly, the improved algorithm stores the alarm sequence in the form of matrix which can enhance the scan efficiency during the candidate stage. Secondly, it adopts the concept of block to organize the candidates since random combinations of candidates can inevitably lead to poor efficiency. Finally, to solve the problem of rules explosion, the algorithm proposes the concept of Simplest Temporal Association Rule and makes use of the block for the production of the rules.The main contribution of this paper is to propose an improved temporal association rule mining algorithm which can promote the efficiency and minimize the quantity of rules produced. Experiments are conducted on real alarm data from a telecommunication corporation to examine the algorithm. The raw data are preprocessed before the experiments. Then, the algorithm is evaluated by the constructed criteria about overall effect and parametic effect. The evaluation of overall effect is to compare the numbers of candidates, frequent sequences and rules as well as the running time. Paremetic evaluations are finished through changing the value of different parameters. The experimental results of the original algorithm and the improved algorithm are compared. Furthemore, the scalability of the improved algorithm is tested by changing the volume of the input data. Eventually, the effectiveness of the improved algorithm is proved.