EBDA: An Effective Bottom-up Discretization Algorithm for Continuous Attributes

Discretization algorithms have played an important role in many areas such as artificial intelligence, data mining and machine learning. In this paper, we propose an effective bottom-up discretization algorithm, namely EBDA. Firstly, we present a novel merging criterion which not only considers the effect of variance on degrees of freedom in the two merged intervals but also the effect of variance on interval difference and data distribution. In addition, we present a new stopping criterion with the aim to control the degree of misclassification and to merge intervals as many as possible. Detailed analysis shows that this algorithm can bring higher accuracy to the discretization process. Empirical experiments on 16 real data sets show that our proposed algorithm generates a better discretization scheme which significantly improves the accuracy of classification than existing algorithms by running C4.5.