Z Khorshidpour
S Hashemi
A. Hamzeh

Measuring similarity or distance plays a key role for data mining and knowledge discovery tasks. A lot of work has been performed on continuous attributes, but for nominal attributes the similarity computation is not relatively well- understood. In this paper, we propose a novel approach to learn a familyof dissimilarity measures for categorical data. Based on these measures distance between two different values of an attribute can be determined by using the certain number of attributes rather than all attributes at once. We evaluate our methods in unsupervised environment, Experiments with real data show that our dissimilarity estimation method improves the accuracy of K-Modes clustering algorithm

Distance function learning, Categorical data, Conditional probability distribution, Kullback Leibler divergence