Back to Research
Model Card
Model Card
Transparent documentation of what this model does, who it is for, and where it should and should not be used.
Model Details
| Model type | XGBoost binary classifier |
| Version | 1.0.0 |
| Trained | 2026 |
| Framework | scikit-learn + XGBoost |
| Input features | 14 BRFSS clinical survey features |
| Output | Binary (0 = No diabetes, 1 = Diabetes) + probability |
| Explainability | SHAP (TreeExplainer) + LIME (LimeTabularExplainer) |
Intended Use
| Primary use | Population-level Type 2 diabetes risk screening |
| Intended users | Healthcare researchers, public health practitioners, students |
| Deployment context | Research and educational demonstration only |
| Out-of-scope use | Clinical diagnosis, individual medical decisions without professional oversight |
Training Data
| Source | CDC Behavioral Risk Factor Surveillance System (BRFSS) |
| Features | 14 demographic, lifestyle, socioeconomic, and clinical variables |
| Target variable | Self-reported diabetes diagnosis (binary) |
| Class balance | Addressed using SMOTE oversampling on training set |
| Train/test split | 80% training, 20% held-out test set (stratified) |
| Cross-validation | 5-fold stratified cross-validation |
Performance
| Accuracy | 0.7208 |
| F1-Score | 0.4484 |
| ROC-AUC | 0.8297 |
| Precision | 0.3123 |
| Recall | 0.7943 |
Limitations
| Population | Training data over-represents certain demographics (BRFSS survey population) |
| Self-reported data | Features are self-reported, introducing potential recall and reporting bias |
| Temporal | Model reflects disease patterns at time of training — may drift over time |
| Geography | Trained on US survey data; may not generalise to other health systems or populations |
| Diagnosis proxy | Target variable is self-reported diagnosis, not laboratory-confirmed |
Ethical Considerations
| Fairness | Model includes race/ethnicity and income as features — SHAP analysis should be monitored for disparate impact across groups |
| Transparency | All predictions include SHAP explanations; no black-box outputs to end users |
| Not a diagnosis | This tool must never be used as a substitute for clinical laboratory testing |
| Data privacy | No patient data is stored beyond the session unless the user is authenticated and opts in |
| Accountability | Predictions should always be reviewed by a qualified healthcare professional before action is taken |
⚠ Important Disclaimer
This model is developed for research and educational purposes as part of a Moringa School capstone project. It is not approved for clinical use. Predictions should never replace clinical laboratory testing or professional medical assessment.