An in-depth comparison of methods handling mixed-attribute data for general fuzzy min-max neural network

A general fuzzy min-max (GFMM) neural network is one of the efficient\nneuro-fuzzy systems for classification problems. However, a disadvantage of\nmost of the current learning algorithms for GFMM is that they can handle\neffectively numerical valued features only. Therefore, this paper provides some\npotential approaches to adapting GFMM learning algorithms for classification\nproblems with mixed-type or only categorical features as they are very common\nin practical applications and often carry very useful information. We will\ncompare and assess three main methods of handling datasets with mixed features,\nincluding the use of encoding methods, the combination of the GFMM model with\nother classifiers, and employing the specific learning algorithms for both\ntypes of features. The experimental results showed that the target and\nJames-Stein are appropriate categorical encoding methods for learning\nalgorithms of GFMM models, while the combination of GFMM neural networks and\ndecision trees is a flexible way to enhance the classification performance of\nGFMM models on datasets with the mixed features. The learning algorithms with\nthe mixed-type feature abilities are potential approaches to deal with\nmixed-attribute data in a natural way, but they need further improvement to\nachieve a better classification accuracy. Based on the analysis, we also\nidentify the strong and weak points of different methods and propose potential\nresearch directions.\n