Enhancing Synthetic Oversampling for Imbalanced Datasets Using Proxima-Orion Neighbors and q-Gaussian Weighting Technique

TL;DR

PO-QG tackles imbalanced data by introducing a two-phase oversampling method that first selects two minority neighbors (Proxima and Orion) using a distance-density based weighting scheme and probabilistic rank-based selection, then synthesizes new samples with a q-Gaussian weighting mechanism to reflect complex minority distributions. The approach emphasizes boundary and sparse-region sampling, balancing the minority class while preserving distributional structure. Extensive experiments on 42 KEEL datasets, eight large UCI datasets, and three real-world sarcopenia-related datasets show consistent improvements in AUC and G-Mean over five established oversampling methods, with Wilcoxon tests confirming statistical significance. The work demonstrates practical potential for improved minority detection in healthcare and related fields, while acknowledging higher computational cost and leaving room for optimization in future work.

Abstract

In this article, we propose a novel oversampling algorithm to increase the number of instances of minority class in an imbalanced dataset. We select two instances, Proxima and Orion, from the set of all minority class instances, based on a combination of relative distance weights and density estimation of majority class instances. Furthermore, the q-Gaussian distribution is used as a weighting mechanism to produce new synthetic instances to improve the representation and diversity. We conduct a comprehensive experiment on 42 datasets extracted from KEEL software and eight datasets from the UCI ML repository to evaluate the usefulness of the proposed (PO-QG) algorithm. Wilcoxon signed-rank test is used to compare the proposed algorithm with five other existing algorithms. The test results show that the proposed technique improves the overall classification performance. We also demonstrate the PO-QG algorithm to a dataset of Indian patients with sarcopenia.

Figures (11)

• Figure 1: Boundary Minority Instances and Their Proximity to Majority Instances
• Figure 2: Flow of PO-QG Algorithm: Synthetic Sample Generation
• Figure 3: Illustration of the PO-QG Algorithm: Synthetic Sample Generation
• Figure 4: Original Data
• Figure 5: Different Oversampling Methods: PO-QG, SMOTE, SMOTE-TL, ADASYN, SMOTE-ENN, and Boundary SMOTE.