A three-stage expert system based on support vector machines for thyroid disease diagnosis

Area of Science:

• Medical Informatics
• Machine Learning in Healthcare
• Computational Biology

Background:

• Thyroid disease diagnosis relies on accurate and efficient methods.
• Existing machine learning approaches for thyroid disease diagnosis have limitations in feature selection and parameter optimization.

Purpose of the Study:

• To develop and evaluate a novel three-stage expert system for thyroid disease diagnosis.
• To enhance diagnostic accuracy through optimized feature selection and Support Vector Machine (SVM) parameter tuning.

Main Methods:

• A three-stage expert system (FS-PSO-SVM) integrating feature selection, SVM classification, and Particle Swarm Optimization (PSO) for parameter optimization.
• Evaluation using a standard thyroid disease dataset and comparison with Grid-SVM and PCA-Grid-SVM methods.
• 10-fold cross-validation (CV) to assess classification accuracy.

Main Results:

• The proposed FS-PSO-SVM system demonstrated superior performance compared to Grid-SVM and PCA-Grid-SVM.
• Achieved the highest reported classification accuracy to date, with a mean accuracy of 97.49% and a maximum of 98.59% via 10-fold CV.
• Significant improvement in discriminative capability through optimized feature subsets and SVM parameters.

Conclusions:

• The FS-PSO-SVM expert system offers a powerful and accurate tool for thyroid disease diagnosis.
• The hybrid approach effectively addresses challenges in feature selection and model optimization.
• This system shows promise for clinical application in improving diagnostic outcomes.