A First Approach To Handle Fuzzy Emerging Patterns Mining On Big Data Problems: The Evaefp-Spark Algorithm

Internet and the new technologies are generating new scenarios with and a significant increase of data volumes. The treatment of this huge quantity of information is impossible with traditional methodologies and we need to design new approaches towards distributed paradigms such as MapReduce. This situation is widely known in the literature as Big Data.This contribution presents a first approach to handle fuzzy emerging patterns in big data environments. This new algorithm is called EvAFP-Spark and is development in Apache Spark based on MapReduce. The use of this paradigm allows us the analysis of huge datasets efficiently. The main idea of EvAEFP-Spark is to modify the methodology of evaluation of the populations in the evolutionary process. In this way, a population is evaluated in the different maps, obtained in the Map phase of the paradigm, and for each one a confusion matrix is obtained. Then, the Reduce function accumulates the confusion matrix for each map in a general matrix in order to evaluate the fitness of the individuals. An experimental study with high dimensional datasets is performed in order to show the advantages of this algorithm in emerging patterns mining.