SOUL: Scala Oversampling and Undersampling Library for imbalance classification

Abstract

The improvements in technology and computation have promoted a global adoption of Data Science. It is devoted to extracting significant knowledge from high amounts of information by means of the application of Artificial Intelligence and Machine Learning tools. Among the different tasks within Data Science, classification is probably the most widespread overall. Focusing on the classification scenario, we often face some datasets in which the number of instances for one of the classes is much lower than that of the remaining ones. This issue is known as the imbalanced classification problem, and it is mainly related to the need for boosting the recognition of the minority class examples. In spite of a large number of solutions that were proposed in the specialized literature to address imbalanced classification, there is a lack of open-source software that compiles the most relevant ones in an easy-to-use and scalable way. In this paper, we present a novel software approach named as SOUL, which stands for Scala Oversampling and Undersampling Library for imbalanced classification. The main capabilities of this new library include a large number of different data preprocessing techniques, efficient execution of these approaches, and a graphical environment to contrast the output for the different preprocessing solutions.