Discretizing Numerical Values by a Fuzzy Clustering Technique

Abstract

The numerical value discretization is an important task of the data preprocessing phase within the intelligent data analysis. This process allows us to reduce the number of values (among other advantages) with which techniques work, reducing the computational cost when it comes to working with large amounts of data. In this paper a numerical value discretization technique is proposed. Specifically, we discretize numerical values using a type of Cauchy distribution obtained from fuzzy clustering technique, being this technique a modification of the well-known Fuzzy C-Means clustering technique. Finally, to test the quality of the membership function we use a neural network technique over several datasets. The results obtained are compared and validated by means of statistical tests, obtaining satisfactory results.