Rule Discovery in Databases with Missing Values Based on Rough Set Model

Abstract

One of the most important problems on rule induction methods is that measures used for rule search will be influenced by missing values. In this paper, a new approach to missing values is introduced, called rough estimation of conditional probabilities. This technique uses three estimation strategies, ground mean, lower and upper methods. Attributes which have missing values will be estimated by these methods and will be checked by constraints for probabilistic rules. The proposed method was evaluated on medical databases, the experimental results of which show that induced rules correctly represented experts’ knowledge.