Distance learning for categorical attribute based on context information

In this paper, we propose a novel method to measure the dissimilarity of categorical data. Our approach is based on two steps, in the first step we select a relevant subset of the whole attributes set that we use as the context for a given attribute and in the second step computes dissimilarity between pair of values of the same attribute using the context defined in the previous step. Dissimilarity between two categorical values of an attribute compute as a combination of dissimilarities between the conditional probability distributions of context attributes given these two values. Experiments with real data show that our dissimilarity estimation method improves the accuracy of the popular nearest neighbor classifier.