A feature selection method using a fuzzy mutual information measure

Attempting to obtain a classifier or a model from datasets could be a cumbersome task, specifically in datasets with a high dimensional datasets. The larger the amount of features the higher the complexity of the problem, and the larger the time expended in generating the outcome -the classifier or the model-. Feature selection has been proved as a good technique for eliminating features that do not add information of the system. There are several different approaches for feature selection, but until our knowledge there are not many different approaches when feature selection is involved with imprecise data and genetic fuzzy systems. In this paper, a feature selection method based on the fuzzy mutual information is proposed. The outlined method is valid either for classifying problems when expertise partitioning is given, and it represents the base of future work including the use of the in case of imprecise data.