A novel convex-hull-based algorithm for classification problems with imbalanced and overlapping data

doi:10.1016/j.eswa.2025.129691

A novel convex-hull-based algorithm for classification problems with imbalanced and overlapping data

2025-09-11

0

PRE

AI

DOI：10.1016/j.eswa.2025.129691

Original

Original request for help

Save

Abstract

En 中文

Classification is a fundamental task in supervised machine learning. This problem becomes challenging when dealing with imbalanced and overlapping datasets. In such cases, learning algorithms often perform well in identifying the labels of majority class data points but exhibit high error rates in predicting the minority class. This paper proposes an innovative method based on the convex-hull concept to enhance classification for imbalanced and overlapping datasets. Unlike undersampling approaches that may lead to the loss of valuable information, our method focuses on preserving the data. The process begins by clustering the data points for each class separately in such a way that no points from the opposite class fall within the convex-hull of each cluster. Then, the support vector machine (SVM) is used to separate every cluster of a given class from the data points of the opposite class. Afterward, data points inside the SVM boundaries are considered as non-overlapping, while those outside the SVM boundaries are identified as overlapping data. The XGBoost algorithm is then employed to classify the data points within the overlapping region. Extensive experiments on a variety of simulated and real-world datasets confirm the effectiveness of the proposed method in terms of various evaluation metrics, compared to existing relevant algorithms for handling imbalanced and overlapping datasets.

A novel convex-hull-based algorithm for classification problems with imbalanced and overlapping data

Abstract

Journal

Organization

Cited Papers