Modeling of voting expert system prototype in classification of acid-base disorders

Measurement of acid-base status and electrolytes in the blood is of great importance since their relationship has a crucial effect on the functioning of all body´s enzyme systems. Therefore, the acid-base status and value of electrolytes are frequently and accurately monitored. The basis of this paper is the modeling of a hybrid fuzzy-neural voting system that will offer diagnosis of diseases incurred on the basis of acid-base disorders as well as anion gap value caused by electrolytes values. The purpose of our paper is to create and compare more fuzzy systems trained by ANFIS method using different membership functions, different modes of fuzzy rule base generation, different learning mechanisms in order to find the best structure in terms of performances and validation results. We have used 264 laboratory data including patients with ten different acid-base disorders and normal patients distributed in the whole sample set. For validation it was used 20 samples which were not used in training process and 10 times 26 samples in cross-validation process. In order to increase reliability, the system is designed on three parallel fuzzy model based on voting principle. The task was carried out using the Matlab interactive environment and its Fuzzy Logic Toolbox with ANFIS method inside it.