Artificial Neural Networks for Diagnosis of Thyroid Disease

摘要

Aritifical neural networks (ANN) have been widely advocated as tools for solving many decision-making problems. In this paper, ANNs is used for the prediction of thyroid disease (TD). Data from expert endocrinologists is used in the training and testing of the ANN algorithm. To speed up the training time faster backpropagation training algorithm is used and the algorithm is implemented on pentium processor. An identification accuracy greater than 80%, a comparable performance to probabiistic and statistical techniques, have been achieved. Furthermore with this algorithm, the training time to achieve such a result is quite less in comparison with the statistical approach, and that the saving in time is substantial. It is seen that th ANN is a fast alternative to classical statistical techniques for prediction and modeing of experimental data. A comparison of the resulting times for backbropagation, faster backpropagation and Levenberg-Marquardt algorithm is presented. From comparison it is seen that faster backpropagation algorithm is most suitable for this application.