A Biomarker-Driven and Interpretable Machine Learning Model for Diagnosing Diabetes Mellitus.

Area of Science:

• Biomedical Informatics
• Computational Biology
• Public Health

Background:

• Diabetes mellitus is a major global health concern, necessitating improved diagnostic tools for timely intervention.
• Current diagnostic methods can be enhanced by leveraging advanced computational approaches for earlier and more accurate detection.

Purpose of the Study:

• To develop and validate machine learning models for predicting diabetes risk using biochemical and physiological data.
• To identify key predictive indicators and assess the performance of different algorithms in diabetes diagnosis.

Main Methods:

• Utilized data from 4335 participants in the National Health and Nutrition Examination Survey (NHANES) database (2017-2020).
• Applied the Boruta algorithm for feature selection, followed by Random Forest (RF), Multi-Layer Perceptron (MLP), and Extreme Gradient Boosting (XGBoost) model development.
• Evaluated model performance using 10-fold cross-validation and testing on a separate dataset, employing metrics like AUC, recall, specificity, and accuracy.

Main Results:

• The Random Forest (RF) model demonstrated superior performance with an Area Under the Curve (AUC) of 0.958, accuracy of 0.913, recall of 0.897, and F1 score of 0.747.
• Feature importance analysis identified glycohemoglobin, glucose, fasting glucose, age, cholesterol, osmolality, BMI, blood urea nitrogen, and insulin as significant predictors.
• SHapley Additive exPlanations (SHAP) and Partial Dependency Plots (PDP) were used for model interpretability.

Conclusions:

• The developed RF model shows significant potential as a supplementary tool for early diabetes diagnosis and risk assessment.
• The findings highlight the utility of machine learning in analyzing complex health data for improved clinical decision-making in diabetes management.