ABSTRACT
A glucose concentration ranging from 72–134 to 111–330 mg/dL is used to represent a non-diabetic/diabetic patient. It is observed that diabetes damages the eyes, kidneys and nerves due to impairment of small blood vessels. Further, hypertension induces kidney failure. A fresh blood drop by piercing fingertip is used to monitor glucose level with a blood glucose meter of a person at home as well as at office. A realistic reusable low-cost non-invasive glucose monitoring system is developed using 802.11a Wi-Fi module. This module measures glucose levels based on the permittivity changes in human blood. The module has 64 subcarriers frequencies spaced with 312.5 kHz bandwidth. Total generated 64 amplitudes and phases are used for channel estimation and in turn to measure blood glucose levels. A Hampel filter suppresses abrupt amplitude variations occurring due to the propagation effects. A principal component analysis (PCA) is carried out on the collected amplitude dataset. The components having minimum eigenvalues is further used to train the model for different glucose concentration using machine learning for accurate detection of diabetes. It also reduces dataset dimension and minimizes the time for training the system. The accuracy obtained is around 90%–95%.
