Identification of Diabetes through Urine Using Gas Sensor and Convolutional Neural Network

Currently, blood glucose disease detection devices are administered invasively by inserting a needle into a finger or a blood vessel. This can cause trauma to the patient, especially if the frequency of examinations is very frequent. Therefore, we need a more comfortable and effective device for early recognition of the signs of diabetes mellitus. Blood sugar monitoring can also be done by smelling the patient's urine. Electronic nose is a system that can be used to identify someone with diseases including diabetes. The electronic nose system used in this study consists of several semiconductor gas sensors. The pattern recognition algorithms utilize convolutional neural network (CNN), learning vector quantization (LVQ), multilayer Perceptron (MLP), Naïve Bayes, k-Nearest Neighbors (k-NN), SVM, PCA-Logistic Regression (PCA-LR), PCA -LVQ, and PCA-MLP. The experimental results show that the CNN has the highest accuracy of 100% in the identification of two classes, namely healthy, and diabetic.