Towards a model for the detection and identification of failures in long haul mobile networks

This work proposes a model to detect and identify failures in Personal Communications Service mobile networks. Among the information available from the network equipment and transiting messages, our investigation concluded that event counters produced by Mobile Switching Centers (MSC) have the best set of attributes to be used in clustering mechanisms. Specifically, Clear Code counters, produced when MSCs establish connections between mobile subscribers, allow the perception and location of failures independently of its geographic region, performance, and subscriber use profiles. To develop our model, Clear Codes collected from 6 MSCs in different country-wide geographic regions are organized in tables to compose the mass of data to be analyzed. The analysis starts by clustering the data into groups using Self-Organizing Maps. Each group is analyzed by an expert that identifies common characteristics and assigns a classification according to the type of faults or behavior each group represents. The resulting classification is evaluated by another expert to demonstrate the model’s abilities. The results showed that the proposed model is able to detect and identify network faults in 5 different categories and evaluate each equipment with respect to its performance. Thus, the information in the output of the model helps the management team to detect faults in the network, identify the faulty MSC, and obtain an overview of the network according to performance.