Multi-Layer Perceptron and Radial Basis Function for Thyroid Disease Diagnosis

Abstract

Thyroid disease is a common health disorder that affect millions of people worldwide. Achieving an accurate diagnosis often involves conducting multiple laboratory tests. However, recent advancements in machine learning have demonstrated promising outcomes in extracting meaningful information from these tests and medical images. By leveraging machine learning techniques, healthcare professionals can enhance their ability to analyze and interpret the data obtained from these tests, leading to more accurate and efficient diagnoses of thyroid dis- eases. This paper proposes a method to enhance thyroid disease diagnosis by combining Multi-Layer Perceptron (MLP) and Radial Basis Function (RBF) neural networks. The proposed method utilizes thyroid-related features to categorize a patient’s status into hyperthyroidism, hypothyroidism, or normal. UCI Machine Learning Repository is used as the training dataset for this study. To ensure high accuracy, data is randomly fitted to both the MLP and RBF networks. Additionally, an adaptive learning rate is applied in the backpropagation training for MLP. This helps avoid local minimums and reduces loss fluctuations and divergence. On the other hand, RBF is trained using the K-means algorithm together with the RLS algorithm. Based on the results, MLP outperforms RBF when the patient’s status is A or B. However, RBF shows better detection when the patient’s status is C. Therefore, the highest accuracy is achieved when the network switches from MLP to RBF based on the patient’s status. Overall, this paper suggests that combining MLP and RBF neural networks can improve the accuracy of thyroid disease diagnosis, especially when considering different patient statuses.