Data Analysis for Developing Blood Glucose Level Control System

Since approximately 10% of people have now diabetes in Japan, the importance of diabetes prevention is increasing. Recently, there are many support programs which allow a person with diabetes to control blood glucose level. However, there are few ways to help non-diabetic people avoid becoming diabetics. Too high peak blood glucose level and prolonged postprandial hyperglycemia can lead to lifestyle-related diseases such as type 2 diabetes. Therefore, it is important to prevent before patients get these diseases. For this purpose, blood glucose level control is required. In this paper, we propose a system for non-diabetic persons to control blood glucose level by predicting it before eating a meal from its image captured. Specifically, we recommend not eating a meal that causes a significant increase in blood glucose level. We analyzed data to create and validate a blood glucose estimation model as the first step toward the realization of a blood glucose level control system. We collected and characterized data on Glycemic Index(GI) of the meal, the time elapsed since the last meal, and the bedtime and sleeping time from four participants to construct a blood glucose level estimation model for each participant using Random Forest.As a result, the constructed estimation models for four participants could estimate blood glucose level with RMSE of 15.41, 12.84, 10, and 10.09, R2 of 0.21, 0.54, 0.75, and 0.82, and finally, MAE of 11.64, 9.232, 6.44, and 6.00.