Predicting Diabetes Mellitus With Machine Learning Techniques.

Area of Science:

• Medical Informatics
• Computational Biology
• Data Science

Background:

• Diabetes mellitus is a chronic hyperglycemia condition with increasing global morbidity.
• Projections indicate 642 million diabetic patients worldwide by 2040, necessitating advanced predictive strategies.
• Machine learning (ML) offers promising applications in medical health, including disease prediction.

Purpose of the Study:

• To evaluate the efficacy of ML algorithms for predicting diabetes mellitus.
• To compare the performance of decision tree, random forest, and neural network models.
• To identify the most accurate ML approach for diabetes prediction using hospital examination data.

Main Methods:

• Utilized hospital physical examination data from Luzhou, China, comprising 14 attributes.
• Applied five-fold cross-validation and independent test experiments for model evaluation.
• Employed dimensionality reduction techniques: Principal Component Analysis (PCA) and Minimum Redundancy Maximum Relevance (mRMR).

Main Results:

• Random forest model achieved the highest prediction accuracy (ACC = 0.8084) when all 14 attributes were utilized.
• The study addressed data imbalance through random data extraction and averaging results over five experiments.
• Model performance was verified through independent test experiments to ensure universal applicability.

Conclusions:

• Machine learning, particularly the random forest algorithm, demonstrates significant potential for accurate diabetes mellitus prediction.
• The findings highlight the effectiveness of using comprehensive hospital examination data for ML-based health predictions.
• Further research can leverage these ML models to aid in early diabetes detection and management.