An Intelligent Neural Network Approach with Comprehensive Optimization to Detect Thyroid

Abstract

Accomplishing a precise analysis for sicknesses remains a considerable task in the arena of health science. This paper proposes a novel approach for analyzing thyroid sicknesses by leveraging Artificial Neural Networks (ANN) to tackle this inherent difficulty. The research is centered on the classification of thyroid diseases as either hyperthyroid or hypothyroid, compared to a normal thyroid, utilizing a comprehensive clinical database. The study meticulously identifies optimal parameters for recommended neural networks (NN), including Multi-Layer Perceptron (MLP), Generalized Feed Forward (GFF) NN, and Jordan Network. Subsequently, three optimal NN models are carefully chosen and subjected to rigorous testing using available datasets. Various data partitioning methods are applied, and extensive experiments are conducted across diverse datasets to showcase the effectiveness and robustness of NN. The MLP NN approach consistently outperforms GFF NN and Jordan Network across various performance metrics, such as accuracy (achieving 100% correct detection with only 0.90% error in multifold cross-validation), training time (0.091 msec), Receiver Operating Curve (ROC area unity) analysis, and mean squared error (0.0031). This comprehensive performance analysis facilitates the identification of the most efficient model for diagnosing thyroid disorders, with MLP NN emerging as the preferred and superior choice.