Preprocessing unbalanced data set based on self-organizing neural networks

Abstract

The unbalanced of data is a common problem in many domains. Using unbalanced data for standard machine learning classifiers significantly affect the obtained performance. In this paper is presented a description of the problem and a review of the main alternatives to solve it. It is also proposed an alternative model, illustrating its application through a case of the medical field. The proposed model manages to get a balanced distribution of instances per class. It is based on the automatic selection of a subset of cases from the majority classes, using the natural groupings of these classes through self-organizing maps. The model is applied to the recognition of heartbeat types and the results are compared with others methods. The results show the feasibility of using this model to address this problem.