Abstract:

The original distance measure of Kmodes clustering algorithm cannot reflect the difference between categorical variables. To overcome this drawback, we propose a new distance measure algorithm based on the intermediate result of Nave Bayes classifier. This algorithm constructs feature vectors to present categorical variables and uses the Euclidean distance of the feature vectors as distance between variables. We implement the Kmodes algorithm with the new derived measure and the experiments on extensive UCI data sets show that the proposal is more effective in comparison with other measure algorithms.

Key words: K-modes clustering algorithm, categorical variables, Nave Bayes classifier, distance measure