GBCD: A Granular-ball Based Discretization Method for Continuous Valued Attributes

Discretization is a wide used method for data preprocessing in data mining and machine learning. However, the existing discretization methods have three drawbacks, including the many parameters and the stopping criterion limited by the data distribution. Meanwhile, most classifiers can only deal with discrete data. In order to solve these problems, a granular-ball based discretization method (GBCD) is proposed for continuous valued attributes. In GBCD, a dataset is granulated into many granular-balls of different sizes. The granular-ball generation is based on k-means algorithm, so the samples in a granular-ball have great similarity. In addition, the center of a granular-ball can express the general feature of the samples in the granular-ball. As a result, the centers of these granular-balls are the discretized samples. The experimental results on UCI benchmark datasets show that GBCD performs better in accuracy than the comparison algorithms.