Combining Feature Space Refinement and Outcome Accumulation Type Evolutionary Rule Discovery

An outcome accumulation-type evolutionary rule discovery method (GNMiner) has been proposed, which has a directed graph structure and is characterized by the accumulation of outcomes acquired in the evolutionary process. Because GNMiner is less efficient at discovery with respect to large feature sets, a feature-selection mechanism is required. In this paper, we propose a framework that integrates evolutionary rule discovery using GNMiner and evolutionary feature selection to refine the search space. The framework is characterized by problem-solving through an evolutionary process in which the two components are interdependent. The experimental results show that the proposed method increases the number of important rules discovered and enhances the fitness value of GNMiner individuals compared with conventional GN-Miner. For feature selection, the results also show that evolutionary computation based on the acquired outcomes of GNMiner is effective.