Among current issues that are challenging to be dealt with in medical science, Diabetes is a noteworthy one. A good number of invasive techniques to measure blood glucose levels have been devised in the past few decades that had been a mixed blessing because it carried multiple limitations along with beneficial sides. Pain, discomfort, and risks of pathogenic infection were some of the basic drawbacks. So, per the demand of time, our main objective of this paper is to introduce a method of blood glucose level measurement that is non-invasive and can overcome the former limitations with better accuracy in a very cost-effective way. Blood glucose concentration can be measured using the PPG (Photoplethysmogram) signal. But to gain higher accuracy we need to consider the physiological variations which lead to erroneous measurement of glucose levels. Using GSR (Galvanic Skin Response) sensor data, these perturbations can be minimized. So, in our model PPG sensor and GSR sensor have been used to accurately measure blood glucose levels. Extracted data from these two sensors as well as recorded blood glucose measurement using conventional glucometer were applied in a deep learning algorithm to measure blood glucose level non-invasively. Then the output of our proposed system was compared to conventional invasive technique. The results showed that our proposed multi sensor-based system improves prediction error of glucose level.
