Modern machine learning approaches for predicting type 2 diabetes mellitus

Between 1990 and 2022, the number of people with diabetes increased from 200 million to 830 million, with the majority of cases attributed to non-insulin-dependent type 2 diabetes. This condition often leads to severe complications such as blindness, kidney failure, myocardial infarction, stroke, and limb amputations. According to WHO estimates, more than 2 million people died in 2021 from diabetes and kidney-related diseases. Global healthcare expenditures related to diabetes were estimated at 966 billion USD in 2021 and are projected to reach 1.054 trillion USD by 2045. Early diagnosis and a personalized approach to treatment can slow disease progression and reduce the risk of vascular complications. At early stages, lifestyle modification alone can achieve long-term normoglycemia without the need for medication. However, nearly half of all people with diabetes worldwide remain undiagnosed. One of the key challenges remains the prevention of hypoglycemia, particularly nocturnal hypoglycemia, which can lead to dangerous complications. Today, the use of continuous glucose monitoring data combined with machine learning algorithms enables the prediction of hypoglycemia risk, taking into account individual patient characteristics. Modern artificial intelligence models rely on large-scale, multimodal data analysis and can adapt in real time. Addressing the issue of data imbalance improves the reliability of predictions. Further progress requires the development of large data libraries and secure systems for information exchange between monitoring devices and clinical applications. The implementation of such technologies enhances treatment effectiveness and improves patients' quality of life. This review highlights the prospects for integrating artificial intelligence into diabetes monitoring and prediction.