DBCS: Density based cluster sampling for solving imbalanced classification problem

Class imbalance of instances is a common problem in the field of data mining and machine learning. A dataset is considered to be imbalanced if one of the classes (further called a minority class or positive class) contains much smaller number of instances than the remaining classes (majority classes or negative class). We describe a new approach to balance data with improved classification. Resampling approaches is a group of popular and flexible techniques to deal with class imbalance, which attempt to resample the original dataset, by oversampling the minority class and/or under sampling the majority class the imbalanced learning problem, there are various methods that solve the imbalanced data at data level and algorithm level. The purpose of this paper is to present a more detailed analysis of density based cluster oversampling and under sampling techniques in terms of density based clustering. The concept of density based cluster sampling called DBCS is to use DBSCAN algorithm for discover clutters and apply oversampling and resampling in minority and majority class respectively. The experimental results of the hybrid DBCS algorithm are compared with an original datasets and other sampling techniques. The results show that the DBCS can improve the predictive performance of the classifiers. In addition, it yields the best in the average of accuracy.