MOEA-EFEP: Multi-Objective Evolutionary Algorithm for Extracting Fuzzy Emerging Patterns

Abstract

Emerging pattern mining is a data mining task that belongs to the supervized descriptive rule discovery framework. Its objective is to find rules that describe emerging behavior or differentiating characteristics with respect to a property of interest. A multiobjective evolutionary algorithm for the extraction of fuzzy emerging patterns (MOEA-EFEP) is described and analyzed in this paper. MOEA-EFEP is the first multi-objective evolutionary algorithm proposed for emerging pattern mining. This approach allows us to get rules whose descriptions of the emerging phenomena are simpler than previous approaches. It is based on the well-known NSGA-II algorithm adapted for the extraction of emerging patterns. The proposal also uses fuzzy logic to deal with numeric variables in order to obtain a knowledge representation close to human reasoning. An experimental study was performed to verify the validity of the proposal. First, it presents a comparison of different rule representations and postprocessing filter strategies, in order to determine an optimal configuration of the proposal. Finally, it is compared with other algorithms for emerging pattern mining in order to determine the quality of the knowledge extracted. The results show that MOEA-EFEP obtains rules with a better description of the emerging or discriminative behavior than other algorithms of the task. The conclusions of this study are supported by the use of statistical tests.