Classification and clustering study in incomplete data domain
Jorge Enrique Rodas Osollo, Freiry Javier Gramajo Lopez · 2001
This report focuses on classification and clustering methods applied to a synthetic dataset, where we gradually increase the missing data and tested---algorithms and tools (C4.5, Roc, AutoClass, Kmeans and K2) for classification and clustering--- independently within each increment. We report preliminary results that suggest that the methods for classification and clustering based on Bayesian or probabilistic structures work well in incomplete data domains and we can intuitively say that it is possible make a kind of Bayesian framework to efficiently share information among its components. These results are not completely satisfactory and we need to test different algorithms. That is, not only classification or clustering, but also other as CBRs for example. This experiment was done because we are motivated about how we can make a general recommendation process description.