Neural networks for fault-prediction in a telecommunications network

The main topic of this paper is fault prediction from large alarm records stored in different databases of non-cooperating network management systems. We have chosen the countrywide data network of Pakistan Telecom (PTCL) as a basis for the investigation of neural network based algorithms to predict faults before they stop a large number of users´ circuits from normal operation. The main problems addressed are the evaluation of alarms, virtual reconstruction of the network and development of tools to overcome the interoperability issues. The motivation behind this work is to assist human operators and minimize the cost of the alarm evaluation process.