Skip to content

The computer network faults classification using a novel hybrid classifier

Student thesis: Doctoral Thesis

  • Karwan Qader
The increasing importance and complexity of networks led to the development of network fault management as a distinct field, providing support for network administrators with quality services and ensuring that networks work appropriately. Fault diagnosis is a central aspect of network fault management. Since faults are unavoidable in communication systems, their quick detection and isolation are essential for the robustness, reliability and accessibility of the system. In large and complex communication networks, automating fault classification is critical. Because of many factors, including the volume of network information, it is hard to solve network fault problems with traditional tools, rendering intelligent analysis a critical method in the process of network fault diagnosis.

This work stated how the common traffic faults in a computer network system could be classified efficiently and improved the performance of the network which is lead to lessening the cost and the time consumed amount. Thus the network administrators will avail to diagnosis the network issues instantly. The experimental work used the datasets of IF-MIB variables which is detected by a researcher in two different scenarios that captured in the router and the server.

To address these issues, this research has conducted several significant contributions, which is relate to the concern of fault management in the computer network system. One of the main tasks of classifying faults is refining the existing datasets. Usually datasets obtained from network analysis tools and experiments are prone to noise, ambiguities and inaccuracies. Thus to produce cleansed datasets that are free from any noises, and inconsistent, two different filtering techniques (normalisation and standard scaling) are proposed.

The optimised fuzzy clustering means (subtractive fuzzy clustering means), which is hybridised with the subtractive clustering is also developed to indicate the optimal number of clusters efficiently. Other researchers did not address the efficiency of their clustering methods.

Further, based on the advantages of the PNN classifier, the SFCM is consolidated with it, and a new proposed classifier model is developed which is defined as a subtractive fuzzy probabilistic neural network classifier (SFPNNC).
Original languageEnglish
Awarding Institution
Supervisors/Advisors
Award dateFeb 2019
Relations Get citation (various referencing formats)

ID: 14426814