.png){kind=icon}
.png){kind=icon}
Return
Certainty-aware support vector machines for fair classification
.png){kind=icon}
PRE
AI
DOI：10.1016/j.neucom.2025.131330.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Support Vector Machine (SVM) is a widely adopted classification algorithm. Many fair machine learning frameworks achieve fairer classification based on SVMs, mitigating systematic bias due to historical discrimination in the training data. However, data imbalance is often inherent in sensitive subgroups, which may cause the learned decision boundary to become skewed toward the majority group, further exacerbating the disadvantages faced by minority groups. To address these fairness concerns, we propose a novel Certainty Factor-based Support Vector Machine (CFSVM) for fair decision-making. CFSVM is an adaptive debiasing framework that incorporates instance-level penalization guided by Certainty Factors (CFs). It dynamically adjusts the influence of individual training samples by estimating local confidence using K-Nearest Neighbors (KNN) and CFs. These scores quantify each sample’s representativeness with respect to both class labels and subgroup distributions, and are used to modulate their contribution to the decision boundary during training. By tailoring the influence of each sample, our proposed CFSVM method mitigates the effects of over-clustering among dominant subgroups and addresses local distributional disparities, especially in scenarios involving imbalanced subgroup representations. We evaluate CFSVM on multiple real-world datasets across several distinct fairness frameworks. Experimental results demonstrate that CFSVM improves fairness metrics while maintaining competitive classification performance, providing a practical and effective solution for fair decision-making in machine learning.
Keywords:
Support Vector Machine
Fairness in Machine Learning
Data Imbalance
Certainty Factors
Adaptive Debiasing
Journal
IF:
6.5
Papers:
2.5W
Citations:
6.5W
Organization
View more organizations