A Self-tuning of Membership Functions for Medical Diagnosis

Abstract

In this paper, a self-tuning of membership functions for fuzzy logic is proposed for medical diagnosis. Our algorithm uses decision tree as a tool to generate three kinds of membership functions which are triangular, bell shape and Gaussian curve. The system can automatically select the best form of membership function for the classification process that can provide the best classification result. The advantage of our system is that it doesn’t need the expert to create membership functions for each feature. But the system can create various membership functions using learning algorithm that learns from the training set. In some domains, user can provide prior knowledge that can be used to enhance the performance of the classifier. However, in medical domain, we found that some diseases are difficult to diagnose. It would not be a problem if that disease has been completely explored in medical area. In order to rule out the patient, we need a domain expert to provide the membership functions for many attributes obtained from the laboratory test. Since the disease has not been completely explored in medical area, the membership function provided by the expert might be biased and lead to the poor classification performance. The performance of our proposed algorithm has been investigated on 2 medical data sets. The experimental results show that our approach can effectively enhance the classification performance compare to neural networks and the traditional fuzzy logic.