Haruna Chiroma - Theories and Applications

This Springer imprint is published by the registered company Springer Nature Switzerland AG

The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland

Computer networks are crucial to todays society as they support communication between individuals, governments, and businesses. They are used not only to connect desktop computers, but also all kinds of electronic devices such as smartphones, wearable devices, sensors, and industrial machines. They also play a key role in emerging domains such as sensor networks [].

To ensure effective and efficient communication between devices, a network must be carefully designed in terms of physical and logical topology, and software must be properly configured. This requires considering various aspects such as budget, facilities, performance, and security requirements. Then, during a networks lifetime, various maintenance tasks must be carried out such as to replace, install, and upgrade equipment and software. Moreover, a key activity is fault management, which is carried out to ensure a networks security, availability, reliability, and optimize its performance [].

Fault management aims at solving problems that are occurring in a network. It consists of four main tasks, which are (1) Detecting, (2) Diagnosing, (3) Isolating, and (4) Fixing network faults [].

Since more then two decades, some attempts at developing computer systems for fault management were made. For instance, in the 1990s, some expert systems were designed that relied on a knowledge base of rules to diagnose network problems. But a drawback of such systems was that specifying rules by hand requires expert knowledge, these rules would not be noise tolerant, and that writing these rules is time consuming and prone to errors [].

To build computer systems for fault management that do not rely heavily on domain experts, a promising fault management approach has been to apply data mining and machine learning-based techniques []. These techniques allow to semi-automatically extract knowledge and learn models from data. Though there has been several studies in this direction, no survey has been published on this topic.