Enhanced glucose forecasting using recurrent neural network and advanced feature engineering

Abstract

The aim of this study is to develop an artificial intelligence (AI)-driven pipeline for forecasting blood glucose levels to mitigate risks associated with hypoglycemia and hyperglycemia. The main research question focuses on the effectiveness of hybrid data preprocessing and feature engineering in enhancing glucose level predictions. The proposed approach employs a hybrid methodology for handling missing data and advanced feature engineering techniques. A recurrent neural network (RNN) model is developed to forecast glucose levels with a lead time of 30 minutes. The model is evaluated using metrics such as root mean square error (RMSE) and mean absolute error (MAE). Experimental results indicate that the proposed pipeline achieves an average RMSE of 19.64 ± 0.11 and an MAE of 13.54 ± 0.11 across all patients. The results demonstrate improved forecasting accuracy, enabling early detection of critical glucose fluctuations. The integration of hybrid preprocessing and RNN modeling effectively predicts glucose levels, providing valuable insights for diabetes management. This approach supports better prevention of glucose emergencies, ultimately enhancing the quality of life for individuals with diabetes.