Managing Uncertainty and Quality in the Classification Process

An important open issue in KDD research is the reveal and the handling of uncertainty. The popular classification approaches do not take into account this feature while they do not exploit properly the significant amount of information included in the results of classification process (i.e., classification scheme), though it will be useful in decision-making. In this paper we present a framework that maintains uncertainty throughout the classification process by maintaining the classification belief and moreover enables assignment of an item to multiple classes with a different belief. Decision support tools are provided for decisions related to: i. relative importance of classes in a data set (i.e., “young vs. old customers”), ii. relative importance of classes across data sets iii. the information content of different data sets. Finally we provide a mechanism for evaluating classification schemes and select the scheme that best fits the data under consideration.