ISSN 2394-3777 (Print) ISSN 2394-3785 (Online) Available online at www.ijartet.com International Journal of Advanced Research Trends in Engineering and Technology (IJARTET) Vol. 3, Issue 9, September 2016 All Rights Reserved © 2016 IJARTET 1 An Efficient Mechanism for Classification of Imbalanced Data Krithika M V 1 , Rajeev Bilagi 2 , Dr. Prashanth C M 3 1 Post Graduate Student, Dept of CS&E, SCE Bangalore, India. 2 Mr.Rajeev Bilagi, Assoc. Prof, Dept of CS&E, SCE Bangalore, India. 3 Dr. Prashanth C M, Prof & HOD, Dept of CS&E, SCE Bangalore, India. Abstract: In many real world applications, there is wide increment in data generation and storage. The classification algorithms are facing a problem in the categorization of highly imbalanced datasets. Classification methods dealt so far centred only on binary class imbalanced problem. All the classification algorithms are biased towards the majority class ignoring most of the significant samples present in the minority class. To resolve this issue, a method called Hybrid Sampling technique is proposed to deal with multi class imbalanced data. The proposed method is an efficient method because it acts by balancing the data distribution of all the classes and imbibes efficient sample selection strategy to undersample the majority class. Experiments are performed using various classifiers and the results of proposed system prove that the classification prediction rate improves when a balanced data having different category of class groupings are considered. Keywords: Classification, data mining, Imbalance Problems, K means Clustering, Multi Class Imbalanced data, Sampling Techniques, Stratified Sampling I. INTRODUCTION Most of the real world applications have to identify the occurrence of rare events from very large datasets. Data mining techniques analyse massive amount of data from various sources and resolve issues of various views by summing them up into useful information [1]. Decision making needs good outlined ways for exploration of data or cognition from various areas. Data mining is the prediction of efficacious information from massive datasets. Classification or categorization plays a pivotal role in the application space of data mining. Classification involves assigning a class label to a set of undefined examples. Classification becomes a serious issue with highly skewed dataset. The classification algorithms proposed so far dealt with imbalanced binary class problem. Evaluating and negotiating the problem of imbalanced data in multiple class domain has been proposed in this study. Class imbalance [2] problem is a predominant issue in the field of data mining and machine learning techniques. All the classification algorithms are biased towards the majority classes, ignoring most of the minority classes that occur very rarely but are found to be the most important. 1.1 Class Imbalanced Data Problem Class imbalance problem is said to occur when the count on collection of samples in one class (superior class) is not less than half the count of the other class (inferior classes). A class that has largest count on the collection items is related as majority class (superior/negative class) and the one that has comparatively less count on the collection items is related as minority (inferior) class or a positive class. As the superior class has large number of training instances, the classifiers show desirable accuracy rates upon observing such class but the categorization rate drops down when an inferior class is observed. Classification algorithms [3] on imbalanced dataset show poor performances due to the following reasons:  The goal of any classification algorithm is to minimize the overall error rates.  They assume the class distribution of different class labels as equal.  Misclassification error rates of all the classes are considered to be equal [4].  Most of the data mining algorithms assume uniform distribution of records among all the classes.