Application of statistical sampling to predict faults from real time alarm data

Abstract

The faults in today's telecommunication systems happen frequently. The reason is simply the complexity of the telecommunication networks and other supported hardware. Various parts of the network may be supplied by various vendors, thus adding to the overall operational complexity. Furthermore these networks may consist of various types of networks, e.g. computer, switching, circuit and wireless networks, inter-operating. The result of such complexities is that a number of faults will happen over unit time intervals. Many of these faults may result in denial of service to the end users, consequently causing revenue losses to the telecommunication companies. Therefore various fault prediction and correction techniques have been proposed. Most, if not all, of these techniques are based on analysis of the historical alarm logs. The Telecom Alarm Sequence Analyzer (TASA) project has proposed associate and episodal rules for historical alarms. We have practically applied TASA episodal rules to identify sequence of alarms and then used the probabilities of these alarm sequences to predict future sequences of alarms. We have used the proposed alarm prediction technique on the real time alarm data of a telecommunication company and predicted future alarms sequences. Furthermore we have compared the predicted alarms against the actual alarm sequences to check the accuracy of our technique. The proposed technique does not make any assumption about the alarm data. We have concluded that the proposed technique has practical usage for alarm prediction in telecommunication networks.