Minority–Majority Mix mean Oversampling Technique: An Efficient Technique to Improve Classification of Imbalanced Data Sets

Abstract

The challenges related to handling of the gigantic imbalanced data volumes are incredible and has set a new trail for its efficient processing. The inventive prospects contained by these huge imbalanced data sets have posed a priority of concern in recent research avenues. The several applications handling imbalanced Big Data sets have noted significance for precise classification while determining unidentified values from these data sets. Traditional classifiers are not able to discourse the imbalance of class distribution among the data samples. A class having fewer samples indicates difficulty in learning, whereas it points to a notable drop in the performance. Recent studies demonstrate that the classifier independent set of oversampling techniques are more capable to efficiently handle the issues raised in imbalanced data sets. An enhanced oversampling technique, viz., Minority–Majority Mix mean Oversampling Technique (MMMmOT), improving classification performance is discussed in detail in this paper. An appropriate consideration of majority as well as minority samples is planned to generate the synthetic samples. The proposed technique is investigated encircling data sets mainly from the UCI repository over Apache Hadoop. Furthermore, the stimulus of maintaining the imbalance ratio with better oversampling instances from the generated pool is analyzed. The results of classification performance are recognized using standard parameters like F-Measure and area under the curve. The achieved experimental outcomes clearly exhibit the preeminence of the presented technique over the traditional techniques.