ENHANCING SEMI-SUPERVISED LEARNING MODELS FOR IMBALANCED CLASS DISTRIBUTION

Abstract

Semi-supervised learning (SSL) has achieved great success in overcoming the difficulties of labeling and making full use of unlabeled data. Semi-supervised learning is an effective approach for addressing the issue of insufficient labeled data, utilizing both labeled and unlabeled datasets. Class imbalance remains a significant challenge, particularly in real-world scenarios. Class dominance imbalance leads to a model bias toward the majority class, hindering the accurate learning and representation of minority classes, which impacts overall  model performance. Current algorithms focus on maximizing overall accuracy but fail to ensure balanced performance across classes. This bias toward the dominant class limits the model's applicability, especially in fields like healthcare, fraud detection, and anomaly detection, where minority class prediction is crucial. This paper proposes novel strategies for semi-supervised learning, specifically targeting imbalanced class distributions. The proposed approach enhances data distribution strategies to address the imbalance issue without compromising model performance. Experimental results demonstrate a significant improvement in the performance of minority classes, validating the effectiveness of the proposed techniques in improving model equity and trustworthiness. This research provides a roadmap for the use of semi-supervised learning in real-life applications where class imbalance is a prevalent issue.