Fuzzy Approach to Data Reliability (original) (raw)

A novel, fuzzy approach to deal with reliability of attribute values in a relational database is presented. The degree of reliability is defined as a fuzzy measure of certainty that the data is correct from user's point of view. The relation scheme is partitioned into a subset of input (completely reliable) and a subset of target (partially reliable) attributes. An information theoretic connectionist network is constructed to evaluate the information content of links between input and target attributes. The network connection weights are used to calculate the reliability degrees of target attribute values. The method is applied to a real-world database, which includes partially reliable information. This work aims at improving reliability of data in a relational database by developing a framework for evaluating and representing reliability of attribute values in database tuples.