HANDLING MISSING ATTRIBUTE VALUES IN DECISION TABLES USING VALUED TOLERANCE APPROACH

Abstract

Rule induction is one of the key areas in data mining as it is applied to a large number of real life data. However, in such real life data, the information is incompletely specified most of the time. To induce rules from these incomplete data, more powerful algorithms are necessary. This research work mainly focuses on a probabilistic approach based on the valued tolerance relation. This thesis is divided into two parts. The first part describes the implementation of the valued tolerance relation. The induced rules are then evaluated based on the error rate due to incorrectly classified and unclassified examples. The second part of this research work shows a comparison of the rules induced by the MLEM2 algorithm that has been implemented before, with the rules induced by the valued tolerance based approach which was implemented as part of this research. Hence, through this thesis, the error rate for the MLEM2 algorithm and the valued tolerance based approach are compared and the results are documented.