Machine learning (ML) and deep learning (DL) approaches have recently attracted considerable attention in the field of diabetes mellitus research because of their potential for predicting and understanding the nuanced facets of the condition. This comprehensive literature review analyzed the present landscape of the predictive modeling of diabetes mellitus. Innovative ML and DL methods were the focus of this study. This study analyzes the persistent problems faced by scientists and engineers working on type 2 diabetes. To systematically summarize the results of 18 carefully selected papers, this study applied the PRISMA method with the addition of methods from the Keele and Durham universities. Several ML and DL methods, including 18 distinct model types, were thoroughly evaluated. Predicting diabetes using tree-based algorithms is highly accurate. In contrast, deep neural networks have shown subpar performance, despite their inherent capacity to handle complicated and large-scale datasets. The importance of feature selection and data balance in enhancing model efficiency was also emphasized in this study as key data preparation methodologies. It has been demonstrated that models trained on structured datasets achieve unprecedented prediction accuracy. This study examined the benefits and drawbacks of numerous machine learning and deep learning approaches. Its purpose is to provide researchers and medical professionals with in-depth knowledge of where diabetes mellitus predictive modeling is at the moment and where it may go in the future. Examination of diabetes mellitus is the main topic of this review.