A Novel Approach to Classify Imbalanced Dataset Based on Rare Attributes and Double Confidences

The major weakness of associative classification is examined. A novel approach for classifying imbalanced dataset is proposed. It is an associative classification. Rules which are un-frequent are used to build the classifier rule set. Besides the confidence of pattern “X→Y”, the confidence of pattern “Y→X” is used in the approach. Further more, only features of rare classes are preserved while training. The good performance of the approach is shown by the experiments.