Internet of Things and Machine Learning–based Diabetes Management System

Effective diabetes management is critical for diabetics which involves continuous glucose level monitoring, generally carried out invasively which is costly, uncomfortable, and inconvenient. Lately, non-invasive approaches have been proposed for low-cost and high-accuracy glucose level monitoring using optical and transdermal methods. This paper presents the design, implementation, and testing of a diabetes management system based on Internet of Things (IOT), machine learning, cloud storage, and mobile application development. The system involves the use of non-invasive glucose level monitoring techniques that use the near infrared spectroscopy (NIRS) and bio-impedance spectroscopy (BIS). It is a painless procedure that involves passing a tiny-amplitude electrical current with a specific frequency through cables connected to electrodes put on the human body. To capture the lowest impedance point, the BIS system executes a frequency sweep. The NIR system employs two distinct light emitting diodes (LEDs) with two distinct wavelengths, each of which has a peak absorption for glucose molecules and is less responsive to other blood components. The machine learning linear regression model is used to estimate the glucose level by using the observed parameters of the two techniques as features, along with some other physical attributes. After estimating the glucose level, the system transmits the information over a Wi-Fi module to a cloud service from where it is accessed for display via a mobile application. The system was tested in terms of the mean absolute error (MAE), mean squared error (MSE), and root-mean squared error (RMSE) and all values are minimal.