Intelligent Diagnosis Method for New Diseases Based on Fuzzy SVM Incremental Learning.

Area of Science:

• Computational biology
• Machine learning in healthcare
• Medical informatics

Background:

• Diagnosing new diseases is difficult due to limited initial case samples, often resulting in low accuracy for intelligent diagnostic systems.
• Traditional Support Vector Machine (SVM) models require retraining on all data for updates, incurring high computational and storage costs and limiting adaptability.

Purpose of the Study:

• To develop a novel disease diagnosis method that overcomes the limitations of standard SVM, particularly for small or evolving datasets.
• To enhance the accuracy and efficiency of intelligent diagnostic systems through an incremental learning approach.

Main Methods:

• Proposed a Fuzzy SVM (FSVM) incremental learning method for disease diagnosis.
• Extracted support vector and boundary sample sets for efficient incremental learning, reducing computational and storage requirements.
• Utilized FSVM to mitigate noise impact from reduced training samples and improve model generalization.

Main Results:

• Improved classification accuracy on the banana dataset from 86.4% to 90.4%.
• Achieved 98.2% diagnostic accuracy for COVID-19, significantly outperforming traditional SVM (84%).
• Demonstrated reduced computational load; with 400 training samples, only 77 were needed for model updates.

Conclusions:

• The proposed Fuzzy SVM incremental learning method offers a more accurate and efficient approach to intelligent disease diagnosis, especially for new or evolving diseases.
• This method effectively handles small sample sizes and reduces the cost of model updating, making it suitable for real-world clinical applications.