Machine learning and deep learning predictive models for type 2 diabetes: a systematic review.

Area of Science:

• Medical Informatics
• Computational Biology
• Data Science

Background:

• Diabetes Mellitus is a chronic disease characterized by elevated blood glucose levels.
• Machine and deep learning are increasingly used for diabetes prediction, yet face challenges in technique selection and model interpretability.

Purpose of the Study:

• To systematically review machine learning techniques for type 2 diabetes prediction.
• To identify optimal models and address challenges in feature transparency and technique heterogeneity.

Main Methods:

• A systematic review following PRISMA and Keele/Durham methodologies.
• Inclusion of 90 studies, extracting model type, complementary techniques, datasets, and performance metrics.
• Comparison of 18 different model types.

Main Results:

• Tree-based algorithms demonstrated superior performance in diabetes prediction.
• Deep Neural Networks were suboptimal despite handling large datasets.
• Data balancing and feature selection techniques significantly improved model efficiency.
• Models trained on tidy datasets achieved near-perfect performance.

Conclusions:

• Tree-based algorithms are recommended for type 2 diabetes predictive modeling.
• Emphasis on data balancing and feature selection is crucial for enhancing model performance.
• Standardization of techniques and transparent feature reporting are needed for improved interpretability.