A systematic review on intelligent diagnosis of diabetes using rule-based machine learning techniques

To improve the diagnostic reliability of diabetes mellitus (DM), rule-based machine learning techniques have been proposed. However, the existing studies are highly diverse with a lack of summarization on the state-of-the-art. To address this gap, we comprehensively reviewed some recent studies. Overall, rule-based methods improved the performance and explainability of the machine learning algorithms, providing direct reference for personalized recommendation and clinical intervention of DM. However, the quality and availability of data limited the reliability of the algorithms. The current algorithms focus on fuzzy system and its optimizations, with a scarce of more complex methods. In the future, the rule-based machine learning algorithms can be improved by using large-scale datasets and more complex structures with better clinical knowledge interpretation, where Internet-of-things and advanced artificial intelligence algorithms will play a key role.