Fuzzy Integral And Fusion Of Multiple Classifiers Applying In Medical Diagnosis

It is difficult for the existing medical technology to diagnose specific diseases such as Parkinson. The starting point of this paper is to find a new effective method of assisting medical diagnosis, and promote this idea to the general case. The general guiding principle of this article is to combine the machine learning method and fuzzy integral with specific medical cases.This paper uses fuzzy integral to structure the model of diagnose disease. The doctors only need to measure several medical treatment indicators of people, import this diagnostic model to get the probability of suffering from the disease. Then the result will assist the doctors'diagnosis. Fuzzy thinking is used in the proposed diagnostic model, and it will use clarity of the figures to express the fuzzy diagnosis. There are two key factors in fuzzy integral. One is the importance of attribute itself. The other is the membership function that the degrees of evidential support for attribute. This article has four main works:Firstly, the method that applies simulated annealing algorithm to multi-category classifier fusion to determine the fuzzy measures is put forward. Moreover, fuzzy integral is compared with Bayesian as fusion operator applying in specific medical cases. Secondly, the method which adopts simulated annealing algorithm and the neural network is proposed to obtain the attribute's fuzzy measures, and use fuzzy integral to medical diagnosis. Thirdly, there are too many attributes in some diseases, and the growth of the number to fuzzy measures is exponentially. In order to simplify the calculation, the Principal Component Analysis is used to get the attribute's fuzzy measures. Finally, how to determine the membership function is another difficult problem. This paper attempts to use mathematical formulas to determine the membership function.At last, we combine the theory with specific medical cases such as Parkinson and Gestational Diabetes Mellitus. The experimental results verify that the methods are feasible. From the experiment, it also shows the value of fuzzy integral and fuzzy integral as a fusion operator to multiple classifiers applying in medical diagnosis.