Feature selection and Ensemble Hierarchical Cluster-based Under-sampling approach for extremely imbalanced datasets: Application to gene classification
Sima Soltani, Javad Sadri, Hassan Ahmadi Torshizi · 2011
Class distribution in many informative datasets is highly imbalance. In high imbalance dataset there are large amount of negative samples and a small part of positives. It is difficult to classify imbalanced datasets. In this paper we propose an Ensemble Hierarchical Cluster-based Under-sampling approach for classification of huge and extremely imbalance datasets. Hierarchical Clustering is used to remove negative samples which are dissimilar to positive samples. Ensemble technique collects results from multiple classifiers to predict class labels. Our experimental results show that our approach is very effective for the classification of extremely imbalanced datasets.