Detection of diabetes from gas analysis of human breath using e-Nose

Diabetes is one of the common disease that many people have suffered especially elderly. However, unfortunately only few of them that aware of this metabolic disease and most of them are undiagnosed. Therefore, in this research we propose low cost, non-invasive, and easy to use system that can distinguish healthy or diabetic patients so they can have early preventive action. A total of 40 e-Nose response signal from breath samples have been collected. There are seven main stages to build this system, the making of e-Nose hardware using microcontroller and gas sensors, ground-truth data acquisitions for the training set, signal processing for denoising using Discrete Wavelet Transform (DWT) and Z-score normalization, statistical features extraction, feature selection for optimization, classification, and e-Nose performance evaluation. The experimental results show that this system can distinguish healthy and diabetes patients with promising performance (95.0% of accuracy, 91.30% precision of diabetes, 94.12% precision of healthy and 0.898 kappa statistic's value) using k-NN classifier.