Moshkovich, H.M., Mechitov, A.I., Co-Authors

Rule induction is aimed at finding stable dependences in data. The approaches to rule induction can be roughly divided into two groups: data-driven and model-driven. Although the majority of induction techniques used in data mining are data driven, implementing some elements of the model-driven approach may be useful in domains where comprehensive prior knowledge exists. This paper illustrates that use of information about ordinal scales for some of the attributes in a classification task may lead to considerable gains in the quality of resulting rule systems. The ordinal classification approach may be used to evaluate how consistent and complete a data set is. Data treatment alternatives are presented to deal with data sets having greater imperfections.