Case Study

MediTrust - Explainable AI Healthcare Risk Prediction Platform

Explainable cardiovascular risk prediction platform using logistic regression, SHAP, FastAPI, PostgreSQL, Docker, and AWS, achieving 0.9665 ROC-AUC.

Logistic RegressionSHAPFastAPIPostgreSQLDockerAWS

Project summary

MediTrust is an explainable AI healthcare risk prediction platform for cardiovascular risk assessment.

The final selected model is Logistic Regression, supported by SHAP explanations and a full-stack deployment architecture.

Dataset

Dataset: UCI Cleveland Heart Disease dataset.

The dataset contains 303 rows and 13 input features, split into 242 training records and 61 test records.

Technical approach

The platform combines a Logistic Regression model, SHAP explainability, a FastAPI backend, PostgreSQL storage, Docker packaging, Gemini-generated summaries, and AWS EC2 deployment.

XGBoost may have been evaluated during experimentation, but it is not presented as the final selected model.

Evaluation

Final selected model: Logistic Regression.

ROC-AUC: 0.9665. Accuracy: 0.8852. F1 Score: 0.8814.

Explainability and deployment

SHAP explainability is used to support interpretable cardiovascular-risk outputs.

The deployment stack includes FastAPI, PostgreSQL, Docker, Gemini-generated summaries, and AWS EC2.

Challenges and limitations

The dataset is small, so the project should be treated as an educational and engineering portfolio system rather than a clinical decision product.