Discretization of Continuous Attributes in Supervised Learning Algorithms

We propose a new algorithm, called CILA, for discretization of continuous attribute. The CILA algorithm can be used with any class labeled data. The tests performed using the CILA algorithm show that it generates discretization schemes with almost always the highest dependence between the class labels and the discrete intervals, and always with significantly lower number of intervals, when compared with other state-of-the-art discretization algorithms. The use of the CILA algorithm as a preprocessing step for a machine learning algorithm significantly improves the results in terms of the accuracy, which are better than by using other discretization algorithms.