Integrating induction and deduction for noisy data mining

Data mining research has been drawing a lot of interest and attention from various fields since late 1980s. The rapid progress has been achieved from three aspects: the prosperity of data mining conferences, the significant number of data mining algorithms, and widely applied areas of data mining techniques. With the continuing growth of the data volumes in many domains, the need of employing data mining techniques provides not only new opportunities but also immense challenges. In this article, we present our study on a challenging topic – integrating induction and deduction for noisy data mining. In particular, we assume the mechanism that corrupts the input data is a set of structured knowledge in the form of Associative Corruption (AC) rules. We apply deductive reasoning to generate the noise corruption rules; make error corrections on the input data with the help of these rules; and perform inductive learning from the corrected input data. Our experimental results show that the proposed integration framework is effective.