期刊名称：International Journal of Computer and Information Technology
印刷版ISSN：2279-0764
出版年度：2014
卷号：3
期号：1
页码：139
出版社：International Journal of Computer and Information Technology
摘要：This paper proposes a new method tolearn time based rules from telecom system alarmdata for prediction of the classes of alarms. A timebased rule associates an alarm class with the StartTimeattribute and other attributes of alarms.The rules are evaluated with the coverage of therules in the training data set. Given a new alarmgenerated at a particular time, its alarm class canbe predicted with a set of time based rules. Wepresent a new algorithm that extracts time basedrules from alarm data through an ant colony optimization(ACO) process. Given an alarm trainingdata, a search space is formulated as a squarematrix indexed by distinctive attribute values.The pheromone at the search space is computedfrom the training data and a time based rule isdiscovered from the pheromone distribution. Thepheromone distribution is updated after a timebased rule is extracted and the search for a newrule starts. A rule pruning process is used to removeredundant rules and increase the predictionaccuracy of the final rule set. We experimentedthe new method on Nokia Simmons (NSN) andEricsson data sets and compared the results ofthe new method and the TimeSeluth system. Thecomparison demonstrated that the new methodoutperformed TimeSeluth in prediction accuracy.
关键词：ant colony optimization; time based;rules; rules discovery; prediction
