Neighborhood-Hypernetwork for Classification of Imbalanced Data

There exists several characteristics in imbalanced dataset, such as classes imbalance, between-class imbalance, overlapping, influenced noise, multi-classification with class imbalance, etc., which will greatly influence the classification performance of algorithms on imbalanced datasets. So far, the model has been widely used on many classification problems, such as DNA microarray data, text classification, stock prediction, and so on. As traditional hypernetwork cannot deal with continuous data directly and will bias to majority class when used on imbalanced data, this paper presents a neighborhood-hypernetwork model for classification of imbalanced data. The paper improves the structure of hypernetwork to make sure it can deal with the issues mentioned. The efficiency and advantage of the proposed approaches are verified by simulation experience on the UCI dataset.