Comparison of Hierarchical and Non-hierarchical Fuzzy Models with Simulation and an Application on Hypertension Data Set

Objective: The aim of this study is to compare the classification performances of hierarchical and non-hierarchical fuzzy models built by using different membership functions. Materials and Methods: In this study, normally distributed data sets containing different number of independent variables (p=3 and p=6) were generated. Besides, the classification performances of hierarchical and non-hierarchical fuzzy models built by using the data set which contained body mass index, fasting blood glucose and triglyceride values of hypertensive (n=206) and control (n=113) people were compared. Results: It was found that there was a significant difference between the fuzzy models (p<0.001). According to the result of both simulation and hypertension data set application, non-hierarchical fuzzy models were found to have better classification performance than hierarchical fuzzy models according to sensitivity, specificity, accuracy and root mean square criteria. Moreover, when number of independent variables was increased, performances of the models increased too and approached to each other. Conclusion: In fuzzy logic methods, data structure, distributions of the variables and correlation between them, how to divide independent variables into categories and which of the fuzzy logic methods is to choose should be examined by taking an expert support.