Discovering Fuzzy Rules with Parallelized Linguistic Variable Elimination

Fuzzy rule discovery in collected data with linguistic variable elimination has been successfully used to create classifiers. These classifiers are composed of a group of fuzzy rules which are interpretable by humans. However, large amounts of collected data cause difficulties to obtain accurate and interpretable fuzzy rules speedily enough. In this paper, several parallelization variations of linguistic variable elimination for fuzzy rule discovery in collected data are presented with the purpose of increasing the speed of the discovery. The parallelization variations are based on parallel programming patterns tailored to the elimination so that the discovery uses several cores of the processor and the parallelization is thought in terms of tasks rather than threads. Four implemented parallel variations are compared with two sequential implementations of the original algorithm for linguistic variable elimination. Experiments show speed improvements with parallelization for various given inputs with different amounts of instances and linguistic variables.