Skip to main content

AI Solutions for MDS: Artificial Intelligence Techniques for Misuse Detection and Localisation in Telecommunication Environments.

Vincent, J., Mintram, R., Phalp, K. T., Anyakoha, C., Bauerdick, H., Gottfried, B. and Muthuraman, S., 2006. AI Solutions for MDS: Artificial Intelligence Techniques for Misuse Detection and Localisation in Telecommunication Environments. Technical Report. EC.

This is the latest version of this eprint.

Full text available as:



Official URL:


This report considers the application of Articial Intelligence (AI) techniques to the problem of misuse detection and misuse localisation within telecommunications environments. A broad survey of techniques is provided, that covers inter alia rule based systems, model-based systems, case based reasoning, pattern matching, clustering and feature extraction, articial neural networks, genetic algorithms, arti cial immune systems, agent based systems, data mining and a variety of hybrid approaches. The report then considers the central issue of event correlation, that is at the heart of many misuse detection and localisation systems. The notion of being able to infer misuse by the correlation of individual temporally distributed events within a multiple data stream environment is explored, and a range of techniques, covering model based approaches, `programmed' AI and machine learning paradigms. It is found that, in general, correlation is best achieved via rule based approaches, but that these suffer from a number of drawbacks, such as the difculty of developing and maintaining an appropriate knowledge base, and the lack of ability to generalise from known misuses to new unseen misuses. Two distinct approaches are evident. One attempts to encode knowledge of known misuses, typically within rules, and use this to screen events. This approach cannot generally detect misuses for which it has not been programmed, i.e. it is prone to issuing false negatives. The other attempts to `learn' the features of event patterns that constitute normal behaviour, and, by observing patterns that do not match expected behaviour, detect when a misuse has occurred. This approach is prone to issuing false positives, i.e. inferring misuse from innocent patterns of behaviour that the system was not trained to recognise. Contemporary approaches are seen to favour hybridisation, often combining detection or localisation mechanisms for both abnormal and normal behaviour, the former to capture known cases of misuse, the latter to capture unknown cases. In some systems, these mechanisms even work together to update each other to increase detection rates and lower false positive rates. It is concluded that hybridisation offers the most promising future direction, but that a rule or state based component is likely to remain, being the most natural approach to the correlation of complex events. The challenge, then, is to mitigate the weaknesses of canonical programmed systems such that learning, generalisation and adaptation are more readily facilitated.

Item Type:Monograph (Technical Report)
Additional Information:"This report is an official deliverable of the EC funded 6th Framework IST project Misuse Detection System (MDS) contract no. 26459. The work presents a comprehensive evaluation of the state-of-the-art in the application of artificial intelligence techniques and other related approaches to the detection and localisation of misuse occurrences within the telecommunications domain. Supported by over €1.3M funding and involving six European partners (including a service provider and equipment supplier) this work has international relevance to the detection of misuses fraud faults and other problems with the growing telecommunications infrastructure. The report led by Bournemouth University was submitted to the Commission in July 2006. This work is particularly important because it offers the most thorough treatment of the subject and makes recommendations for future MDS developments that are being carried through by the project and are expected to influence for example next generation fraud detection systems for mobile phone networks."
Group:Faculty of Science & Technology
ID Code:11353
Deposited By: Dr Keith Phalp
Deposited On:15 Sep 2009 19:21
Last Modified:14 Mar 2022 13:25

Available Versions of this Item


Downloads per month over past year

More statistics for this item...
Repository Staff Only -