Enhanced AdaBoost Algorithm with Modified Weighting Scheme for Imbalanced Problems

Abstract

Class imbalance is notable critical issues arise during classification of particular applications datasets which is real world. This would leads to added misclassification in specifically one class that compares to other classes. This issue called class imbalance would occurs if one class possess large amount of records and in other hand alternative class (es) possess specifically little number of the records. In order to deal with class imbalance problem, various approaches are introduced and carried out. These methods are primarily based on under sampling the records in majority class or over sampling the minority class records. However, both approaches will produce over fitting or under fitting problems for the trained classifier. This paper deals with the modification of adaboost to enhance the classification performance. Class reweight strategy in adaboost is modified based on percentage of misclassification in previous iteration.