Worldwide, diabetes imposes significant health pressure on humans due to a group of metabolic disorders. Diabetes is regarded as a comprehensive health issue owing to the proliferation of diabetic patients globally. To deplete the chance of extreme complications with diabetes and improve the treatment of diabetes, early diagnosis of diabetes is necessary.
  Machine learning and artificial intelligence currently conduct the early detection and diagnosis of diabetes via automated learning concepts and provide more advantageous decisive support than manual diagnosis. Moreover, many research articles have focused on machine learning-based techniques for automatic diabetes detection, diagnosis, and management in recent times.

Various Machine Learning Methods: Machine learning algorithms create diabetes detection and diagnosis and classification models for recognizing interesting patterns for diagnosing and treating it. Some of the machine learning algorithms are listed below:
• K-nearest neighbor (KNN) classifier - Several variants of the KNN algorithm to classify diabetes based on similar classes of diabetes exist nearby. Recently, diabetes type 1 has been classified using KNN.
• Naïve Bayes (NB) – NB identifies diabetes based on the count of feature occurrences. The medical record is also utilized to predict the patient-s degree of risk and danger level possibilities.
• Support Vector Machine (SVM) – SVM categorizes diabetes data as linear and non-linear. SVM provides good precision in detecting diabetes utilized with electrocardiogram data.
• Linear Discriminant Analysis (LDA) – LDA supports multiclass classification. Diabetes diagnosis enabled with LDA used for rigorous diagnosis of type 2 diabetes.
• Decision Trees (DT) – DT imparts a high precision score for early diagnosis of diabetes type 2 enabled with other tree models such as a coarse, medium, and fine tree.
• Random forest (RF) – RF-based feature extraction is utilized for detecting diabetes type 2 with better accuracy.
• Deep learning (DL) – DL are exploited for applications related to diabetes detection and evolved to discriminate huge amount of diabetes data and also find patterns from unrecognized diabetes data.
• Ensemble learning classifiers - Three types of ensemble classifiers are recently applied to diagnose diabetes type 2; namely, the bagged tree (EBaT), boosted tree (EBoT), and subspace KNN(SKNN) with better precision.
• Unsupervised classifiers – K means clustering is employed to detect type 2 diabetes and pre-diabetes with optimal performance.

Future Scopes and Confrontations of Machine Learning in Diagnosis of Diabetes: Even though many methods and algorithms have been developed in machine learning ordeep learning in diabetes diagnosis, some challenges remain unsolved and need better consideration in future perspectives.
  •  Over-fitting of the machine learning models for diabetes detection caused the problem of small datasets and the inability to deal with new data. Thus, a feasible machine learning model for handling such a problem needs to be implemented.
  •  The absence of repeatability and external validation is another challenge in machine learning for diagnosing and predicting diabetes.
  •  Machine learning models are restricted to handling raw data. Comparatively, deep learning models provide better performance in such cases.
  •  A combined model with supervised, unsupervised, and reinforcement learning concepts to be developed to understand diabetic diagnosis better and implement personalized medical care.
  •  In the future, the integration of deep learning, artificial intelligence, and cloud computing in diagnosing diabetes will be investigated for consistent support in healthcare.
  •  Researchers currently explore machine learning to build smart devices for diabetes management systems.