Detection and Prediction of Diabetes Using Simple Fuzzy-Perceptron Learning Network

Abstract

The fuzzy inference system with expert knowledge can propose interpretable solutions for the uncertainties of clinic data. The learning concept of perceptron networks is simple and close to human thinking. In this study, we used fuzzy inference systems and perceptron learning networks (FIS-PLN) to detect and predict diabetes. For diagnosis of diabetes, insulin, glucose, and BMI are critical and relevant indices. In the detection system, the medical data of insulin, glucose, and BMI were sent to the fuzzy system in advance before training the PLN. The fuzzy system inferred a cross-effect grade that revealed the impact of the medical features on diabetes. The cross-effect grade and other medical data were combined and applied to train the PLN. The testing results demonstrated that under the same simulation conditions and medical features, the FIS-PLN model performed better predictions than PLN. The prediction accuracy approached 79.4% and the AUC of the FIS-PLN model was near 0.843.