Hybrid Data Clustering Based on Dependency Structure and Gibbs Sampling
A new method for data clustering is presented in this paper. It can cluster data set with both continuous and discrete data effectively. By using this method, the values of cluster variable are viewed as missing data. At first, the missing data are initialized randomly. All those data are revised through the iteration by combining Gibbs sampling with the dependency structure that is built according to prior knowledge or built as star-shaped structure alternatively. A penalty coefficient is introduced to extend the MDL scoring function and the optimal cluster number is determined by using the extended MDL scoring function and the statistical methods.
KeywordsDependency Structure Gibbs Sampling Cluster Number Cluster Variable Cluster Accuracy
Unable to display preview. Download preview PDF.
- 1.Chen, S.M., Hsiao, H.R.: A New Method to Estimate Null Values in Relational Database Systems Based on Automatic Clustering Techniques. Information Sciences: an International Journal 69, 1–2 (2005)Google Scholar
- 2.Cheeseman, P., Kelly, J., Self, M., Stutz, J., Taylor, W., Freeman, D.: AutoClass: A Bayesian Classification System. In: Laird, J. (ed.) Proceedings of the 15th International Conference on Machine Learning, pp. 54–64. Morgan Kaufmann, San Mateo (1988)Google Scholar
- 3.Cheeseman, P., Stutz, J.: Bayesian Classification (AutoClass): Theory and Results. In: Fayyad, U.M., Piatetsky-Shapiro, G., Smyth, P., Uthurusamy, R. (eds.), pp. 153–180. AAAI/MIT Press, Cambridge (1996)Google Scholar
- 5.Mao, S.S., Wang, J.L., Pu, X.L.: Advanced Mathematical Statistics, 1st edn., pp. 401–459. China Higher Education Press, Beijing, Springer, Berlin (1998)Google Scholar
- 7.Domingos, P., Pazzani, M.: On the Optimality of the Simple Bayesian Classifier Under Zero-one Loss. Machine Learning 130, 2–3 (1997)Google Scholar
- 8.Murphy, S.L., Aha, D.W.: UCI Repository of Machine Learning Databases, http://www.ics.uci.edu/~mlearn/MLRepository.html