Cluster-based sampling of multiclass imbalanced data

Abstract

The aim of this paper is to improve the classification performance based on the multiclass imbalanced datasets. In this paper, we introduce a new resampling approach based on Clustering with sampling for Multiclass Imbalanced classification using Ensemble (C-MIEN). C-MIEN uses the clustering approach to create a new training set for each cluster. The new training sets consist of the new label of instances with similar characteristics. This step is applied to reduce the number of classes then the complexity problem can be easily solved by C-MIEN. After that, we apply two resampling techniques (oversampling and undersampling) to rebalance the class distribution. Finally, the class distribution of each training set is balanced and ensemble approaches are used to combine the models obtained with the proposed method through majority vote. Moreover, we carefully design the experiments and analyze the behavior of C-MIEN with different parameters (imbalance ratio and number of classifiers). The experimental results show that C-MIEN achieved higher performance than state-of-the-art methods. This paper is concerned with improving the classification performance on multiclass imbalanced dataset, which is even more complicated. Moreover, the higher degree of class imbalance may increase the difficulty of multiclass classification. Solutions for two-class problems are not directly applicable to multiclass cases. One of the famous methods is decomposition technique, which is decompose the multiclass dataset into a series of binary classification problems and then use a two-class learner for a classification task [13,18,32,34] such as One-Against-One (OAO) [60], One-Against-All (OAA) [7]. Several decomposition methods use ensemble approach to combine the models obtained from the binary class classifiers. However, using decomposition with sampling technique is not practical for this problem because it is time consuming. Moreover, in case OAA, results of each class label assignment are not comparable due to the decision can be made differently for different classes [54]. Hence, reducing the number of classes and comparing labels becomes a key issue for applying the resampling technique in multiclass cases.In this paper, we develop a resampling algorithm for multiclass imbalance problem based on clustering approach namely C-MIEN. Firstly, k-means is used to split the set of instances into two clusters. For each cluster, hybird sampling methods are used. Then, final training sets (classes are balanced) are used to build an emsemble. Finally, the prediction is obtained by combining the results from both clusters through a majority vote. C-MIEN is an extension of our previous works [42][43][44] that focused on different classifiers in the classification part. In our previous works [42,43], we did not apply ensemble in the classification part. The re-balancing process was different from this paper. Moreover, we carefully design the experiments and analyze the behavior of C-MIEN with different parameters (imbalance ratio and number of c...