There are numerous mathematical models of the glucose homeostasis system, serving as vital tools in understanding the pathways and treatment strategies for diabetes. This paper provides an extensive review of these models, categorising them into clinical and non-clinical frameworks. We explore various approaches, including minimal models, minimal control networks, and the more recent data-driven models. Through our review, we highlight the strengths and limitations of each category, identifying significant gaps in the existing literature. A trend is the increasing importance of data-driven, machine learning methods, which capture the complex, nonlinear dynamics of glucose regulation more effectively than traditional mathematical models. These advanced methods offer the potential for more personalised and adaptive treatment strategies, reflecting the variability in patient responses and disease progression. Furthermore, machine learning models can continuously improve and adapt with the influx of new data, making them highly dynamic and responsive to changes over time. This adaptability contrasts with the static nature of many traditional models, allowing for real-time adjustments in treatment plans. By leveraging diverse heterogeneous datasets and sophisticated algorithms, machine learning approaches can uncover hidden patterns and relationships that might be overlooked by conventional models. Our findings emphasise the need for further research and development in this area, aiming to replace traditional mathematical models with more sophisticated, data-driven approaches. By addressing these gaps, we contribute to the ongoing efforts to improve diabetes management and treatment outcomes, leading to more accurate and individualised treatment interventions.