A method for Handling Inconsistencies in Rule-based Classifiers

Researches on distributed data mining have as the main interest the development of algorithms and approaches that make possible the analysis of large and physically distributed datasets proposing better solutions in terms of costs and computational complexity. In such a scenario, the handling of inconsistent classification rules, generated from distributed data sets is an important task, because inconsistencies can compromise the performance of the classifiers. The aim of this work is the development of a new methodology for knowledge integration based on paraconsistent logic. The method is capable to take reliable decisions in occurrences of inconsistent rules generated from a set of rule-based classifiers. The advantage of the proposed method is to avoid high flow of messages and data between distributed processors throughout the rules analysis generated from distributed data sets, because only local information is used for the generation of a global ruleset. Then, paraconsistent logic is used as a certainty management approach to infer the class for a test example. We could observe in the experiments done over different datasets that the method generates quite good results.