Deep Learning-Based Imbalanced Classification With Fuzzy Support Vector Machine.

Area of Science:

• Machine Learning
• Data Science
• Biomedical Informatics

Background:

• Imbalanced classification is a common challenge in medical diagnosis, biomedicine, and smart city applications.
• Traditional methods struggle with imbalanced data, leading to bias towards majority classes and poor minority class recognition.

Purpose of the Study:

• To propose a novel imbalanced classification method, DFSVM, combining deep learning and fuzzy support vector machines.
• To improve classification performance on imbalanced datasets, particularly in minority class identification.

Main Methods:

• DFSVM utilizes a deep neural network with triplet loss for data embedding, enhancing intra-class similarity and inter-class separability.
• Oversampling in the embedding space using a feature and center distance method generates diverse samples and prevents overfitting.
• A cost-sensitive fuzzy support vector machine (FSVM) is employed as the final classifier to prioritize minority class accuracy.

Main Results:

• Experimental results on biological and real-world datasets demonstrate the effectiveness of the DFSVM method.
• DFSVM achieved promising classification performance, outperforming traditional approaches on imbalanced data.

Conclusions:

• The proposed DFSVM method offers a robust solution for imbalanced classification problems.
• DFSVM shows significant potential for applications requiring accurate minority class detection in imbalanced datasets.