Blood Glucose Level Prediction in Type 1 Diabetes Using Machine Learning

Soon Jynn Chu, Nalaka Amarasiri, Sandesh Giri, Priyata Kafle · 2025

Read Blood Glucose Level Prediction in Type 1 Diabetes Using Machine Learning by Soon Jynn Chu, Nalaka Amarasiri, Sandesh Giri, Priyata Kafle (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2502-00065.txt.

Introduction

Chu, Amarasiri, Giri, and Kafle predict blood glucose in type one diabetes with machine learning in twenty twenty-five. The disease destroys insulin making cells, so patients dose insulin daily. Continuous monitors give near real time readings to guide those doses. Forecasting the next reading helps patients act before problems start.

The team trained several model types on the recent DiaTrend data set. Approaches included neural networks, reinforcement learning, and combined regressors. Predictions targeted glucose thirty minutes ahead. Models were compared across different blood sugar ranges with standard metrics. Results show how each method performs under varied conditions. Code is released openly for other researchers.

Catalog record

CREATORSoon Jynn Chu, Nalaka Amarasiri, Sandesh Giri, Priyata Kafle
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE27 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2502-00065

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Blood Glucose Level Prediction in Type 1 Diabetes Using Machine Learning Soon Jynn Chu, Nalaka Amarasiri, Sandesh Giri, Priyata Kafle Abstract Type 1 Diabetes is a chronic autoimmune condition in which the immune system attacks and destroys insulin-producing beta cells in the pancreas, resulting in little to no insulin production. Insulin helps glucose in your blood enter your muscle, fat, and liver cells so they can use it for energy or store it for later use. If insulin is insufficient, it causes sugar to build up in the blood and leads to serious health problems. People with Type 1 Diabetes need synthetic insulin every day. In diabetes management, continuous glucose monitoring is an important feature that provides near real-time blood glucose data. It is useful in deciding the synthetic insulin dose. In this research work, we used machine learning tools, deep neural networks, deep reinforcement learning, and voting and stacking regressors to predict blood glucose levels at 30-min time intervals using the latest DiaTrend dataset. Predicting blood glucose levels is useful in better diabetes management systems. The trained models were compared using several evaluation metrics. Our evaluation results demonstrate the performance of various models across different glycemic conditions for blood glucose prediction. The source codes of this work can be found in: https://github.com/soon-jynn-chu/t1d_bg_prediction 1 Introduction Type 1 Diabetes (T1D) is a chronic autoimmune condition characterized by the destruction of insulin-producing beta cells in the pancreas, resulting in little to no endogenous insulin production [ 1 ] . This deficiency leads to dysregulation of blood glucose levels, necessitating lifelong external insulin administration to maintain normoglycemia [ 2 ] . T1D affects millions of people worldwide, with its incidence increasing by approximately 3-4% annually, particularly in younger populations [ 3 ] . The management of T1D is complex and demanding, requiring constant vigilance to balance insulin dosing with various factors that influence blood glucose levels, such as food intake, physical activity, stress, and illness [ 4 ] . The primary goal of T1D management is to maintain blood glucose levels within a target range to prevent both short-term complications and long-term microvascular and macrovascular complications [ 5 ] . Understanding the various glycemic values is crucial for effective T1D management. Blood glucose levels are categorized as hypoglycemia when below 70 mg/dL (3.9 mmol/L), within the target range of 70-180 mg/dL (3.9-10.0 mmol/L), and hyperglycemia when above 180 mg/dL (10.0 mmol/L). Hemoglobin A1C (HbA1c), which indicates average blood glucose levels over the past 2-3 months, has a target of less than 7.0% for most adults with T1D. Time In Range (TIR) is another important metric, intending to spend more than 70% of time within the target range [ 6 ] . Glycemic variability, measured by metrics such as standard deviation or the coefficient of variation, is also significant, with lower variability associated with better outcomes [ 7 ] . In recent years, technological advancements have revolutionized diabetes management, particularly through the development of Continuous Glucose Monitoring (CGM) systems. CGM devices are wearable sensors that measure interstitial glucose levels at regular intervals, typically every 5 minutes, providing near real-time glucose data to the user [ 8 ] . This continuous stream of information offers several advantages over traditional fingerstick blood glucose measurements, including improved glycemic control through more frequent glucose readings, early detection and prevention of hypo- and hyperglycemic events, identification of glucose trends and patterns, and enhanced decision-making for insulin dosing and lifestyle choices [ 9 ] . CGM systems provide a wealth of data that can be interpreted to improve diabetes management. The current glucose reading displays the most recent interstitial glucose measurement, updated every 5 minutes for most systems [ 10 ] . Trend arrows indicate the direction and rate of glucose change, helping predict short-term glucose fluctuations. The Ambulatory Glucose Profile (AGP) offers a visualization of glucose data over time, aiding in the identification of patterns and trends [ 11 ] . The Glucose Management Indicator (GMI) estimates HbA1c based on average CGM glucose readings, providing a more immediate indicator of glycemic control. Additionally, customizable alerts and alarms for high and low glucose levels, including predictive alerts for impending hypo- or hyperglycemia, enhance the user’s ability to manage their condition proactively [ 12 ] . In the context of advancing diabetes management technologies, the DiaTrend [ 13 ] dataset has emerged as an invaluable resource for researchers and data scientists. This comprehensive dataset comprises intensive longitudinal data from wearable medical devices, including 27,561 days of CGM data and 8,220 days of insulin pump data from 54 patients with T1D, aged 19-74 years, with 17 males and 37 females. The richness and scope of the DiaTrend dataset provide an unprecedented opportunity to develop and validate Machine Learning (ML) models for various aspects of diabetes care, including blood glucose prediction, detection of adverse glycemic events, optimization of insulin delivery, discovery of digital biomarkers for glycemic control, and analysis of adherence patterns to wearable medical devices. Our research leverages the DiaTrend dataset to develop and compare various ML methods for predicting blood glucose levels using CGM data in 30-minute intervals. We explore a range of algorithms, including Support Vector Regressor (SVR), Light Gradient Boosting Machine (LGB), Multi-Layer Perceptron (MLP), Long Short-Term Memory (LSTM) networks, Gated Recurrent Units (GRU), Deep Deterministic Policy Gradient (DDPG), Twin Delayed Deep Deterministic Policy Gradient (TD3), and Soft Actor-Critic (SAC). By evaluating these diverse approaches, we aim to identify the most effective methods for accurate insulin level prediction. This research has the potential to significantly enhance diabetes management strategies and contribute to the development of more sophisticated closed-loop systems. The potential impact of this research extends beyond immediate improvements in glycemic control. Accurate prediction of blood glucose levels can lead to more precise insulin dosing recommendations, reduced risk of hypoglycemia and hyperglycemia, improved overall quality of life for individuals with T1D, advancements in personalized medicine approaches for diabetes management, and development of more robust artificial pancreas systems. Furthermore, the insights gained from this research could inform the design of next-generation CGM systems and insulin pumps, potentially leading to enhanced prediction algorithms integrated directly into diabetes management devices, improved decision support tools for healthcare providers and patients, and more effective strategies for managing challenging scenarios such as exercise, illness, and stress. 2 Related Work Recent advances in ML and Artificial Intelligence (AI) have spurred the development of innovative models for predicting blood glucose levels, optimizing insulin delivery, and improving diabetes management, particularly for individuals with T1D. The related works discussed in this section address various aspects of T1D care, including adverse event prediction, insulin dosing, and personalized treatment approaches, leveraging both novel datasets and advanced modeling techniques. Zheng et al. [ 14 ] proposed BG-BERT, a self-supervised learning framework designed to predict blood glucose levels with a specific focus on adverse events such as hyperglycemia and hypoglycemia, which are often underrepresented in datasets. BG-BERT employs a masked autoencoder to capture contextual information from blood glucose records and uses the Synthetic Minority Oversampling Technique (SMOTE), a data augmentation technique, to enhance sensitivity to adverse events. Evaluated against benchmark datasets, BG-BERT demonstrated a 9.5% improvement in prediction accuracy and a 44.9% increase in sensitivity to adverse events compared to state-of-the-art models. This

Read the complete text