An effective noise reduction technique for class imbalance classification

The paper presents a unique approach to handle noisy instances in the data sources using the novel technique of priority instance picking for weak range feature subsets. The technique used in the proposed approach quickly identifies the noisy instances in the data source than the benchmark C4.5 algorithm. The C4.5 algorithm also removes the noisy instances from the formed decision tree but in the final stage by applying the pruning technique. The results conducted on 12 UCI datasets suggest that the proposed approach performs better than the benchmark algorithm.