Decision Trellis Models for Tuple Categorization in Databases

We introduce a probabilistic graphical model for supervised learning on databases with categorical attributes. The proposed graph contains hidden variables that play a role similar to nodes in decision trees and each of their states either corresponds to a class label or to a single attribute test. As a major diierence with respect to decision trees, the selection of the attribute to be tested is probabilistic. Thus, the architecture can be used to assess the probability that a tuple belongs to some class, given the predictive attributes. The training algorithm can be easily derived in the general framework of graphical models, using expectation-maximization (EM) for nding the optimal parameters. We propose decision trellises as an alternative to decision trees in the context of tuple categorization in databases, which is an important step for building data mining systems. Preliminary experiments on some standard databases are reported, comparing the classiication accuracy of decision trellises and decision trees induced by C4.5.