-TSVM: A Robust Transductive Support Vector Machine and its Application to the Detection of COVID-19 Infected...

Area of Science:

• Machine Learning
• Data Science
• Medical Imaging Analysis

Background:

• Training machine learning models with incomplete or noisy data presents significant challenges.
• Existing models often struggle with datasets containing both missing labels and label noise.
• Semi-supervised learning methods are crucial for leveraging partially labeled data.

Purpose of the Study:

• To develop a robust machine learning model capable of handling datasets with missing labels and label noise.
• To enhance the performance of Transductive Support Vector Machines (TSVM) in challenging data scenarios.
• To apply the developed model for the detection of coronavirus from chest X-ray images.

Main Methods:

• Proposed a novel approach named Pi-TSVM, integrating a truncated pinball loss function with TSVM for noise robustness.
• Derived both primal and dual formulations for the robust TSVM, supporting linear and non-linear kernels.
• Conducted experiments on synthetic and real-world datasets, including small and large-scale data.

Main Results:

• Demonstrated superior robustness of Pi-TSVM compared to existing methods in the presence of label noise.
• Showcased the model's capability to effectively train on noisy data and accurately identify missing labels.
• Achieved successful detection of coronavirus patients using chest X-ray images with the Pi-TSVM model.

Conclusions:

• Pi-TSVM offers a robust solution for semi-supervised learning on datasets with missing labels and label noise.
• The model's ability to handle noisy data and infer missing labels has practical implications in medical diagnostics.
• The successful application in coronavirus detection highlights the potential of Pi-TSVM in real-world medical imaging challenges.