Interactive exploration of interesting findings in the Telecommunication Network Alarm Sequence Analyzer TASA

Ln this paper we describe the final version of a knowledge discovery system , Telecommunication Network Alarm Sequence Analyzer (TASA), for telecommuni cation networks alarm data analysis. The system is based on the discovery o f recurrent, temporal patterns of alarms in databases; these patterns, epis ode rules, can be used in the construction of real-time alarm correlation s ystems. Also association rules are used for identifying relationships betwe en alarm properties. TASA uses a methodology for knowledge discovery in dat abases (KDD) where one first discovers large collections of patterns at onc e, and then performs interactive retrievals from the collection of patterns . The proposed methodology suits very well such KDD formalisms as associati on and episode rules, where large collections of potentially interesting ru les can be found efficiently. When searching for the most interesting rules : simple threshold-like restrictions, such as rule frequency and confidence may satisfy a large number of rules. In TASA, this problem can be alleviat ed by templates and pattern expressions that describe the form of rules tha t are to be selected or rejected. Using templates the user can flexibly spe cify the focus of interest, and also iteratively refine it. Different versi ons of TASA have been in prototype use in four telecommunication companies since the beginning of 1995. TASA has been found useful in, e.g. finding lo ng-term, rather frequently occurring dependencies, creating an overview of a short-term alarm sequence, and evaluating the alarm data base consistency and correctness. (C) 1999 Elsevier Science B.V. All rights reserved.