Predicting Blood Glucose Levels in Type 1 Diabetes Using Deep Learning and Regression Techniques
Jagendra Singh, Brijendra Pratap Singh, Monika Dandotiya, Pongkit Ekvitayavetchanukul, Manoj Kumar Rana, Bakshish Singh
Abstract
Accurate blood glucose prediction is very critical for the effective management of Type 1 Diabetes as it allows timely interventions and prevents fluctuations in blood sugar levels. Due to such shortcomings, the traditional methods of monitoring that are followed are consistently unable to provide real-time insights and often cannot inform the patients about the stable levels of their glucose. This research studies advanced deep learning applications based on Long Short-Term Memory (LSTM), Gated recurrent unit (GRU), Autoregressive Integrated Moving Average (ARIMA), and linear regression models to predict blood glucose levels in terms of insulin intake dosage, consumption of carbohydrates, level of physical activity, and sleep patterns. The model performances with respect to Mean Absolute Error (MAE), Root Mean Square Error (RMSE), Mean absolute percentage error (MAPE), and accuracy are deployed for training and performance evaluation. Results were obtained, which show that LSTM performed well in comparison to others with an accuracy of 97.68%, followed by ARIMA (92.23%), GRU (91.23%), and Linear Regression at an accuracy of 89.45%. Real-time testing also gave the results consistent with the above, showing accuracy at close par with the glucose values. Therefore, LSTM in diabetes management systems may allow for blood glucose monitoring in real-time and customised intervention. This research contributes towards developing intelligent systems with the capacity to predict problems and enhance patient outcomes and quality of life for those with Type 1 Diabetes by embedding deep learning models into predictive accuracy and timeliness.