A Rule Induction Approach to Forecasting Critical Alarms in a Telecommunication Network

Abstract

This paper proposes a white box method of predicting critical alarms so they can be mitigated and understood by engineers. Forecasting these alarms will avoid outages and maintain the agreed service level which is beneficial to both the provider of telecommunication services and the consumers. The paper evaluates several item set mining approaches on a set of alarms of the British Telecom (BT) national telecommunication network and proposes a novel transformation of the data to enable the discovery of patterns undetectable by current item set mining approaches. The result is a method for rule induction that predicts alarms with high precision using a wide range of features.

A Frequent Pattern Conjunction Heuristic for Rule Generation in Data Streams

“…The Motivation for developing FGRI is the compelling need to unlock information of actionable value in a number of potential applications for which currently no data stream mining algorithms can be applied directly, some of which are outlined below. For example, techniques to predict network alarms and bottlenecks are employed in the UK national telecommunication network, however, the utilised methods often do not enable the tracing of the causality of such alarms [4]. A descriptive rule-set consisting of rules such as the above could enable applications to identify causality of alarms in telecommunication networks.…”