# Machine Learning based prediction of Glucose Levels in Type 1 Diabetes Patients with the use of Continuous Glucose Monitoring Data

**Authors:** Jakub J. Dylag  
**arXiv:** [2302.12856](https://arxiv.org/abs/2302.12856)  
**Date:** 2023-02-24  
**Summary page:** https://www.summarizepaper.com/en/paper/2302.12856v1/

## Key points

- The study focuses on using Machine Learning (ML) models to predict glucose levels in patients with Type 1 Diabetes, using Continuous Glucose Monitoring (CGM) data.
- CGM devices offer non-intrusive and real-time insights into a patient's blood glucose concentrations, which is crucial for diabetes management.
- Three ML models were used: Linear Regression, Hidden Markov Model, and Long-Short Term Memory Network.
- By leveraging a patient's past 11 hours of blood glucose concentration measurements, the models predict the glucose level for the next 60 minutes.
- Performance metrics used were Root Mean Squared Error (RMSE), normalised energy of the second-order differences (ESOD), and F1 score.
- Python programming language was used along with various libraries such as Jupyter Notebooks, PyTorch, hmmlearn, Scikit-learn, Numpy, Pandas, Matplotlib, Seaborn and Pickle to develop their ML models.
- CITY dataset published by Jaeb Center for Health Research was used to train their models.
- An optimal training methodology for the CITY dataset was established that may be leveraged by future model development.
- Results were aligned with similar state-of-the-art ML models with LSTM having RMSE of 28.55; however no significant advantage was observed over classical Auto-regressive AR models.
- Compelling insights into LSTM prediction behaviour could increase public and legislative trust and understanding progressing certification of ML models in Artificial Pancreas Systems (APS).
- Overall this study demonstrates that ML-based prediction approaches can provide substantial quality-of-life improvements for diabetes patients while offering vital tools for monitoring diabetes.

## Summary

This study focuses on the use of Machine Learning (ML) models for predicting glucose levels in patients with Type 1 Diabetes, using Continuous Glucose Monitoring (CGM) data. The prevention of hypo/hyperglycemic events is a crucial aspect of diabetes management, and CGM devices offer non-intrusive and real-time insights into a patient's blood glucose concentrations. The study implements a regression-based prediction approach recursively, using three ML models: Linear Regression, Hidden Markov Model, and Long-Short Term Memory Network. By leveraging a patient's past 11 hours of blood glucose concentration measurements, the models predict the glucose level for the next 60 minutes. The performance of the models is assessed using Root Mean Squared Error (RMSE), normalised energy of the second-order differences (ESOD), and F1 score as performance metrics. The researchers used Python programming language along with various libraries such as Jupyter Notebooks, PyTorch, hmmlearn, Scikit-learn, Numpy, Pandas, Matplotlib, Seaborn and Pickle to develop their ML models. They also used CITY dataset published by Jaeb Center for Health Research to train their models. The research team established an optimal training methodology for the CITY dataset that may be leveraged by future model development. The results were aligned with similar state-of-the-art ML models with LSTM having RMSE of 28.55; however no significant advantage was observed over classical Auto-regressive AR models. Compelling insights into LSTM prediction behaviour could increase public and legislative trust and understanding progressing certification of ML models in Artificial Pancreas Systems (APS). Overall this study demonstrates that ML-based prediction approaches can provide substantial quality-of-life improvements for diabetes patients while offering vital tools for monitoring diabetes.