Machine Learning in Diabetes Modeling

Machine learning (ML) has become an integral component of diabetes research. Its applications encompass prediction, identification, classification, and diagnosis of diabetes-related conditions. To comprehensively explore the real-world implications of machine learning in diabetes, it is imperative to initiate the publication of an efficient survey article from computer science perspective. A standardized database tailored for systematic reviews has been established, encompassing a substantial collection of 1150 articles. Consequently, the predominant objective of this study is to introduce an enhanced PRISMA-based methodology for reviewing and evaluating the performance of machine learning in diabetes contexts. The results of this research cannot be used for the purpose of meta-analysis. Instead, its suitable to have an insight on general performance of ML models which generally outperform other ML methods. The results indicate the most pupular and effective ML in diabetes research. The findings underscore accuracy as the most reliable and comprehensive yardstick for assessing machine learning techniques across various diabetes applications. Furthermore, artificial neural networks (ANN) and random forests (RF) exhibit exceptional suitability, particularly when juxtaposed against alternative machine learning approaches. The majority of investigations predominantly align with detection or estimation pursuits. Additionally, we advocate for the integration of hybrid machine learning techniques in ongoing diabetes research endeavors.