PROJECT TITLE :

Minority Sub-region Estimation-based Oversampling for Imbalance Learning

ABSTRACT:

One of the challenges that has arisen in recent years is a problem known as class imbalance, which is characterized by a skew distribution that favors the majority. Several different oversampling strategies have been proposed as a solution to this issue. Some of these oversampling strategies combine the oversampling procedure with the Clustering algorithm, which ensures that newly generated synthetic samples will be grouped together. Nevertheless, due to a property of the clustering algorithm itself, dispersed samples that share the same minority sub-region but come from different geographic locations tend to be divided into distinct groups. As a result, the procedure for oversampling that follows is carried out the majority of the time in incomplete minority subregions where synthetic samples do not adequately cover the entire minority region. In addition, to the best of our knowledge, none of the currently available algorithms is designed to directly estimate minority sub-regions with regard to the class imbalance problem. First of all, a brand new grouping algorithm that has been given the name Direction Distribution-based Minority Sub-region Estimation (DDMSE) has been proposed. The new algorithm makes use of the intuitive observation that minorities with the same sub-region almost always distribute in the same direction when compared to other majorities. This allows the new algorithm to estimate minority sub-regions while deftly ignoring negative impacts brought on by the distance factor, similar to how clustering algorithms do. Last but not least, in those minority sub-regions, brand new synthetic samples are produced. The experimental results on real-world datasets show that the performance of this oversampling method is comparable to that of other cutting-edge oversampling methods.