A semi-supervised learning based method: Laplacian support vector machine used in diabetes disease diagnosis

Area of Science:

• Computational biology
• Machine learning in healthcare
• Pattern recognition for disease diagnosis

Background:

• Disease diagnosis relies heavily on accurate pattern recognition.
• Semi-supervised learning offers potential for improving diagnostic accuracy, particularly when labeled data is scarce.
• The Pima Indians diabetes dataset provides a benchmark for evaluating predictive models.

Purpose of the Study:

• To evaluate the effectiveness of Laplacian support vector machine (LapSVM) for diabetes disease prediction.
• To compare the performance of LapSVM in fully-supervised versus semi-supervised learning paradigms.
• To assess LapSVM's utility in scenarios with limited class-labeled data.

Main Methods:

• Utilized the Pima Indians diabetes dataset, comprising female patients of Pima Indian heritage aged 21+.
• Implemented LapSVM as a fully-supervised learning classifier for diabetes prediction.
• Implemented LapSVM as a semi-supervised learning classifier for diabetes prediction.

Main Results:

• Fully-supervised LapSVM achieved an accuracy of 79.17% in diabetes prediction.
• Semi-supervised LapSVM achieved a higher prediction accuracy of 82.29%.
• The achieved accuracy of 82.29% surpasses previous reported results for this dataset.

Conclusions:

• Laplacian support vector machine (LapSVM) is a promising method for diabetes disease prediction.
• Semi-supervised learning with LapSVM can outperform fully-supervised approaches, especially with limited labeled data.
• LapSVM offers a valuable tool for physicians in diabetes diagnosis, particularly in data-scarce situations.