Machine/deep Learning Predictive Techniques For Blood Glucose Level

Allowing diabetic patients to predict their blood glucose level (BGL) is an important task for self-management of their metabolic disease. This allows to avoid hypo or hyperglycaemia by taking appropriate actions. Currently, this is possible due to the development of machine learning (ML) and Deep Learning (DL) techniques which are successfully used in many prediction tasks. This paper evaluates and compares the performances of six ML/DL techniques to forecast BGL predictions; four deep learning techniques: CNN, LSTM, GRU, and two ML/statistic techniques: SVR and AR. The evaluation of the performance of the six regressors were in term of four criteria: Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Mean Magnitude Relative Error (MMRE), and Predictive Level (PRED). In addition, the Scott-Knott (SK) were used to evaluate the statistical significance test and to rank the regressors. The results show that AR was the best for 5 minutes ahead forecasting with a mean of RMSE equal to 8.67 mg/dl.