Intelligent Heart Disease Risk Assessment System Based on Machine Learning: From Model Construction to Production Deployment

Heart-Disease-Prediction is a production-grade machine learning application designed to provide intelligent heart disease risk assessment. The project implements a complete workflow from data preprocessing and model training to containerized deployment. Key highlights include: using FastAPI to provide high-performance API services, building an interactive dashboard with Streamlit, achieving model interpretability via SHAP, and managing experiments and models with MLflow. The system has a prediction accuracy of 88.5%, offering a convenient risk assessment tool for medical institutions and individual users.

Cardiovascular disease is one of the leading causes of death globally, and early risk identification is crucial for preventive healthcare. Traditional assessments rely on doctors' judgments and limited indicators, while machine learning provides new possibilities for large-scale, automated, and high-precision assessments. Addressing this need, this project not only achieves a prediction accuracy of 88.5% but also transforms the model into a deployable, interpretable, and user-friendly practical application, facilitating risk assessment in medical scenarios.

The project adopts modern machine learning engineering best practices, with a tech stack including:

• Backend Service: FastAPI to build RESTful APIs, supporting high concurrency and automatically generating OpenAPI documentation to simplify integration;
• Frontend Interaction: Streamlit to quickly build interactive visual dashboards without deep frontend technical knowledge;
• Model Interpretability: Integrate SHAP values to quantify feature contributions, making model predictions transparent and trustworthy;
• Experiment and Model Management: MLflow to track experiment parameters and version-control models;
• Deployment: Docker containerization ensures environment consistency, simplifying deployment and migration.

The system's core functions include:

1. Multi-dimensional Input: Supports indicators such as age, gender, blood pressure, cholesterol, chest pain type, exercise ECG results, and lifestyle habits;
2. Real-time Assessment: Returns risk predictions in milliseconds based on the trained model, with an 88.5% accuracy that can serve as a clinical auxiliary reference;
3. Personalized Analysis: Uses SHAP visualization to show the impact of each indicator on risk (elevating or protective effects), compares personal indicators with average levels, and helps users improve their health in a targeted manner.

Highlights of the project's engineering practice:

• End-to-End MLOps Workflow: Covers the complete lifecycle from data exploration, feature engineering, model training, evaluation and optimization to deployment monitoring;
• Interpretability First: Integrates SHAP as a core function to meet the demand for model transparency in medical scenarios;
• Modular Code: Separates modules such as data preprocessing, model definition, API services, and frontend interfaces for easy maintenance and expansion.

The project's application scenarios include:

• Personal Health Management: Users assess risks regularly and adjust their lifestyles based on SHAP analysis;
• Clinical Auxiliary Decision-Making: Doctors quickly screen high-risk patients, allocate resources, and explain conditions using SHAP results;
• Medical Education and Research: Serves as a teaching case or for researchers to secondary-develop and explore other disease areas.

This project successfully transforms machine learning into a practical medical auxiliary tool, covering core elements such as accuracy, interpretability, deployability, and maintainability. In the future, with the accumulation of medical data and advances in algorithms, such tools will play a greater role in preventive healthcare, and the open-source nature also provides possibilities for community contributions and function expansion.