Systematic Evaluation of Machine Learning-Based Predictive Model for Diabetes

The paper aims to systematically evaluate the prediction model for diabetes based on machine learning (ML). It conducted literature research in Baidu Scholar, National Library of China, Web of Weipu, CNKI, and Wanfang databases for literature on prediction models for diabetes constructed by ML, and the search period spanned from January 2018 to February 2023. The paper completed literature screening and data extraction independently and used predictive models to construct a research data extraction and quality evaluation checklist (CHARMS) to evaluate the quality of the included literature and screened high-quality literature for discussion. Results: A total of 13 high-quality studies were collected, including 5 ML models, with an area under the ROC curve ranging from 0.720 to 0.97. Laboratory indicators such as age, BMI, blood sugar concentration, waistline, and diabetes history are the main predictive factors. Conclusions: Diabetes prediction models constructed using ML can accurately identify the risk of diabetes, and their predictive performance is superior to traditional risk prediction models. The available literature on the topic exhibits a low overall risk of bias, however, the applicability level of the prediction model is considered average.