MOO-Points – Distance-based Method for Multi-objective Optimization in the Imbalanced Data Classification Task

Real-world applications of machine learning often require a dedicated approach sensitive to different aspects of processing quality depending on the application and domain. Additionally, the goal cannot be reduced to a single metric that does not consider, for example, different error costs for different classes. Thus, multi-objective optimization could be used. The described problem often goes hand in hand with difficult data, such as imbalanced data. This paper proposes the MOO-Points method dedicated to imbalanced data classification task in multi-objective optimization. It is based on distance methods, thus enables the ability to adjust the parameters. The article compares its performance with state-of-the-art methods using synthetic and real data.