Prediction of Diabetes Among Homeless Adults Using Artificial Intelligence: Suggested Recommendations

Area of Science:

• Public Health
• Medical Informatics
• Artificial Intelligence

Background:

• Diabetes mellitus presents a significant global health burden, disproportionately affecting vulnerable populations such as the homeless.
• Early diabetes prediction is crucial for reducing healthcare costs and enhancing intervention effectiveness.
• Identifying specific predictors within the homeless population is vital for targeted prevention strategies.

Purpose of the Study:

• To identify key predictors of diabetes mellitus among homeless adults.
• To develop and evaluate artificial intelligence (AI) models for early diabetes prediction in this demographic.
• To provide data-driven recommendations for diabetes prevention initiatives.

Main Methods:

• A case-control study involving 150 homeless adults in Giza, Egypt, analyzing 43 variables.
• Feature selection techniques reduced predictors to 13 key variables, including physiological measures and lifestyle factors.
• A stacking ensemble model with XGBoost was developed and evaluated using cross-validation, outperforming individual classifiers.

Main Results:

• Key predictors identified include BMI, systolic blood pressure, waist circumference, lifestyle, comorbidities, and age.
• Individual machine learning models showed moderate predictive performance.
• The stacking ensemble model achieved high accuracy (95.45%), precision (100%), recall (93.75%), F1-score (0.968), and AUC-ROC (0.979).

Conclusions:

• Machine learning models, particularly ensemble methods, demonstrate high reliability in predicting diabetes.
• The proposed hybrid stacking model offers superior predictive performance compared to conventional classifiers, especially with imbalanced medical data.
• Integrating AI-powered diagnostic tools into clinical practice is recommended for early diabetes detection and management in at-risk populations.