Deep learning based thyroid prediction with opposition learning based red panda optimization feature selection.

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Biomedical Data Analysis

Background:

• Thyroid disease prediction requires accurate and robust models.
• Existing methods may suffer from biased outcomes and suboptimal feature extraction.
• Publicly available datasets offer potential for developing advanced predictive tools.

Purpose of the Study:

• To introduce a novel, cascaded autoencoder-based recurrent model for thyroid prediction.
• To develop an Opposition Learning-based Red Panda Optimization (OL_RPO) algorithm for optimal feature selection.
• To enhance thyroid prediction accuracy and robustness using an Enhanced Transformer Model.

Main Methods:

• Preprocessing three public datasets for standardization and balance.
• Utilizing a cascaded autoencoder-simple recurrent model for spatio-temporal feature extraction.
• Employing the OL_RPO algorithm for optimal feature selection.
• Performing thyroid prediction with an Enhanced Transformer Model.

Main Results:

• The proposed model achieved high performance metrics: Accuracy (99%), Specificity (99.2%), Sensitivity (99.01%), F-Score (98.501%), PPV (98.1%), and NPV (1.9%).
• A very low error rate of 0.07689 was recorded.
• The OL_RPO algorithm effectively enhanced predictive model efficiency.

Conclusions:

• The novel approach demonstrates significant potential for accurate and robust thyroid prediction.
• The integration of advanced AI models and optimization algorithms offers a promising direction for medical diagnostics.
• The proposed method provides a standardized and balanced approach to thyroid prediction using public data.