Mapping Rules Based Data Mining for Effective Decision Support Application

Due to the learning problem on skewed distribution of data sets, such as data sets of credit card fraud detection, which tend to produce poor predictive accuracy over the minority class by traditional machine learning algorithms, mapping rules based data mining approach (MRDMA) is proposed in this paper to make effective classification decision support on the minority class. MRDMA constructs suitable information granules (IGs) by fuzzy ART, and then hierarchical clustering analysis is employed to produce mapping rules from IGs to final classes. When new inputted data clustered to the IGs by continue on-line learning of fuzzy ART, the final class can soon be decided by the mapping rules. The experimental results show that MRDMA has better classification performance on skewed data sets than SVM and C4.5.