imbalanced-learn-extra: A Python Package for Novel Oversampling Algorithms

Learning from imbalanced data is a common challenge in supervised learning, as most classifiers assume balanced class distributions. Among the strategies to mitigate this issue, oversampling algorithms offer a flexible and model-agnostic solution by generating synthetic samples for minority classes. In this paper, we introduce the imbalanced-learn-extra Python library, an open-source extension of the imbalanced-learn ecosystem that provides additional oversampling techniques for research and practical use. The library integrates seamlessly with Scikit-Learn, allowing users to easily incorporate it into existing workflows. It implements Geometric SMOTE, a geometrically enhanced drop-in replacement for the original SMOTE algorithm, and clustering-based oversampling methods such as KMeans-SMOTE and G-SOMO, which combine existing imbalanced-learn oversamplers with Scikit-Learn clustering algorithms to address within-class imbalances. Rather than re-assessing the performance of these algorithms, which has already been thoroughly evaluated in prior studies, this paper focuses on their software design, implementation, and practical use within a unified framework.