Model Optimization in Imbalanced Regression

Imbalanced domain learning aims to produce accurate models in predicting instances that, though underrepresented, are of utmost importance for the domain. Research this field has been mainly focused on classification tasks. Comparatively, number studies carried out context regression tasks is negligible. One main reasons lack loss functions capable focusing minimizing errors extreme (rare) values. Recently, an evaluation metric was introduced: Squared Error Relevance Area (SERA). This posits a bigger emphasis committed at values while also accounting performance overall target variable domain, thus preventing severe bias. However, its effectiveness as optimization unknown. In paper, our goal study impacts using SERA criterion imbalanced Using gradient boosting algorithms proof concept, we perform experimental with 36 data sets different domains and sizes. Results show that used objective function practically better than produced by their respective standard prediction confirms can be embedded into optimization-based scenarios.