Mining rules from a database according to multiple measures of interest

Research output: Contribution to conferencePaper


In data mining, nugget discovery is the discovery of interesting rules that apply to a target class. There are a number of algorithms specifically designed for the extraction of nuggets. These algorithms generally use some pre-defined measure of the interest of a nugget and search the space of all possible nuggets looking for the most interesting nugget according to the defined measure. Many measures of interest may be defined on a rule. For example accuracy, coverage and simplicity are three such measures. Finding the best nuggets according to a set of interest measures mounts to a multi-objective optimisation problem. In previous research, heuristic methods (Genetic algorithms, Simulated Annealing and Tabu Search) have been used to optimise a single measure of interest which was defined to combine a number of important properties of a nugget. This approach, although successful, is sometimes limiting in nugget discovery as the user may want to vary the criteria of the search (i.e. the interest measure used) and investigate the nuggets produced. This paper proposes to use multi objective optimisation heuristic techniques to allow the user to interactively select a number of interest measures (complimentary or conflicting) and deliver the best nuggets (the pareto-optimal set) according to those measures.
Original languageEnglish
Publication statusPublished - Nov 2002
EventMOMH Multiple Objective Metaheuristic Workshop - Paris, France
Duration: 4 Nov 20025 Nov 2002


ConferenceMOMH Multiple Objective Metaheuristic Workshop

Cite this