ZDF: An adaptive discretization method to improve separability of classes
Fatemeh Kaveh-Yazdy, Mohammad-Reza Zare-Mirakabad · 2013
Distance-based classification methods use distances between un-labeled samples and labeled samples to assign a class label to it. In a problem space, the distance is computed based on attributes with different values. In continuous space, different values reflect different conditions and arise from problem's nature, while differences in discretized datasets may be occurred as a result of discretization process. Thus, we propose a discretization method for optimal cut point place selection to minimize the inter-class distances. Experimental results show that the performance of our method in increasing the separability of samples in classes is significantly better than the results of entropy-based discretization.