Optimizing Thyroid Disease Diagnosis Using a Hybrid Random Forest and MLP Approach

Identifying thyroid dysfunction is crucial for maintaining human health, due to the thyroid gland's role in controlling metabolism, hormonal balance, and overall physiological functions. Traditional diagnostic methods, although beneficial, can require considerable time and are prone to human error. Machine learning (ML) techniques provide a feasible solution by automating diagnostic processes and improving classification accuracy. This study examines eight machine learning algorithms—Support Vector Machines (SVM), Random Forest (RF), Decision Tree (DT), Naive Bayes (NB), Logistic Regression (LR), K-Nearest Neighbors (KNN), Multi-Layer Perceptron (MLP), and Linear Discriminant Analysis (LDA)—for diagnosing thyroid disease, classifying it into three categories: hyperthyroidism, hypothyroidism, and normal. A hybrid algorithm combining RF and MLP was developed to enhance performance. The hybrid model employs RF for feature representation and MLP for complex pattern classification. The assessment metrics—accuracy, precision, recall, and F1-score—demonstrated the hybrid model's superiority over individual methods, achieving an accuracy of 99.15%, precision of 99.05%, recall of 99.10%, and F1-score of 99.07%. The findings underscore the potential of hybrid machine learning methodologies to improve traditional diagnostic practices, reduce errors, and streamline clinical workflows.