Thyroid disease detection using enhanced extreme learning machine based on drop-connect method.

Area of Science:

• Endocrinology
• Computer Science
• Machine Learning

Background:

• Thyroid disorders are significant endocrine diseases with long-term physiological effects.
• Current machine learning methods struggle with reliable multi-class thyroid disease diagnosis.
• Overfitting and generalization issues persist in traditional Extreme Learning Machine (ELM) models.

Purpose of the Study:

• To develop an Enhanced Extreme Learning Machine (EELM) with Drop-Connect regularization for improved thyroid disease classification.
• To evaluate the EELM's performance in a clinically relevant four-class diagnostic scenario.
• To enhance generalization and reduce overfitting in thyroid disease diagnostic models.

Main Methods:

• A seven-step framework including data preprocessing, model building, training, and evaluation.
• Implementation of Drop-Connect regularization within the Enhanced Extreme Learning Machine (EELM).
• Assessment on a unified four-class thyroid classification task (hypothyroidism, hyperthyroidism, sick-euthyroid, normal) using 10-fold cross-validation.

Main Results:

• The EELM achieved an average accuracy of approximately 82% for multi-class classification.
• Up to 99.89% accuracy was reached in binary classification tasks, demonstrating effective discrimination.
• Statistical validation using ANOVA and paired t-tests confirmed significant improvements (p < 0.05) over baseline models.

Conclusions:

• The proposed EELM offers a clinically applicable and statistically supported method for thyroid disease classification.
• The EELM demonstrates robust and reliable multi-class diagnostic performance.
• This approach provides a computationally effective solution for endocrine disorder diagnosis.