Recursive Ensemble Approach for Incremental Learning of Non-Stationary Imbalanced Data
Pradnya A.Jain, Roshani Raut, P. R. Deshmukh · International Journal of Computer Applications · 2014
Learning non-stationary data stream is much difficult as many real world data mining applications involve learning from imbalanced data sets.Imbalance dataset consist of data having minority and majority classes.Classifiers have high productivity accuracy on majority classes and Low productivity accuracy on minority classes.Imbalanced class partition over data stream demands a technique to intensify the underrepresented class concepts for increased overall performance.To alleviate the challenges brought by these problems, this paper propose the recursive ensemble approach (REA).This approach reduces problem of imbalance data by learning minority and majority instances arrived at incremental time.In Practical analysis REA results are compare with Synthetic Minority Over-sampling Technique (SMOTE) and predicted results proves that REA gives better performance as compare to SMOTE on synthetic and real time datasets.