Fuzzy System Modeling Based On Distance Combined Data

At present,the characteristics of the data are varied.To fully tap such data become the focus of many research,the characteristics of the data sometimes affect the nature of the data itself,such as medical data,many features are closely linked,but some of the characteristics of no contact,put together all the characteristics of research between the loss of data information,and even make wrong judgment.Therefore,the data are first grouped into feature groups,and different feature groups are given different distance measurements,so the hidden information between the data can be used to improve the accuracy of the classifier.The work of this paper mainly includes the following two aspects.In this paper,a novel Takagi-Sugeno(T-S)fuzzy system based on an improved fuzzy clustering algorithm is developed to fully leverage useful information in medical data.By combining the classical Fuzzy C-Means(FCM)with adaptive parameters in distance measures and simultaneously keeping the basic structure of T-S fuzzy systems,the proposed fuzzy system has its adaptive modelling superiority in prediction for medical data.The experimental results about the adopted medical datasets indicate both promising performance and feasibility of the proposed fuzzy system.Based on the classical fuzzy C-Means clustering method,a new classifier design method is proposed.First,an alternate clustering method based on distance combination data is developed to be seen as a promotion of the concept of conditional fuzzy clustering.It has some known prior prototypes and can also process data that can be grouped by features.A special method for initializing clustering centers is presented,then the proposed clustering method is used to construct the logical precursors of the Takagi-Sugeno fuzzy system based on the IF-THEN rule,and then the error function is minimized using the gradient descent method.The information between data is fully mined by adjusting the impact factor during the clustering process to make the clustering better and thus improve the accuracy of the classifier.Finally,a large number of experimental analyses of real data sets that can be feature-grouped were performed to verify the effectiveness of the classifier,and the experimental results showed that the improved algorithm has higher accuracy and feasibility.