HealthPredict -- Clinical Risk Prediction Web App¶

Status: Live demo Demo: health.robiriu-dev.my.id (demo login: demo@healthpredict.demo / demo1234)

Executive Summary¶

A self-hostable web app that turns a handful of routine clinical values into an instant, explained risk estimate for common conditions. Each condition is scored by a Logistic Regression and a Random Forest, ensembled, and every result breaks down exactly which values drove the score up or down -- so it reads as a screening aid, not a black box.

Built in a clean, separated architecture (clinical config, ML training, prediction layer and web layer are independent), with email/password auth, a results-history dashboard, an admin console, and a light/dark interface.

Conditions¶

Diabetes, Heart Disease, Chronic Kidney Disease, Liver Disorder and Thyroid Disorder, each with its own clinically-realistic feature set. The same framework extends to additional conditions (Parkinson's, cancers) by adding a model.

How it works¶

clinical values ─► Logistic Regression ┐
                                        ├─ ensemble probability ─► risk band + explained factors
                   Random Forest ───────┘

• Two models per condition -- a Logistic Regression and a Random Forest, trained per condition; their probabilities are averaged for a steadier estimate.
• Explainable -- the Logistic Regression's coefficients rank each input's contribution; the result page shows the share each factor took and flags values outside the healthy range.
• Trustworthy -- the app never invents numbers; the score comes from the trained models on the values you enter.

Key Features¶

• Per-condition input forms with healthy-range hints
• Risk gauge, model-agreement panel (LogReg / RF / AUC) and ranked factor breakdown
• Results-history dashboard with trend and distribution charts
• Email/password authentication; admin view of all users and predictions
• Light / dark theme
• SQLite for development via SQLAlchemy ORM -- one environment variable switches it to MySQL for production

Technology Stack¶

Layer	Technology
Backend	Python, Flask, SQLAlchemy, Flask-Login
Machine learning	scikit-learn -- Logistic Regression + Random Forest, joblib-persisted
Frontend	Server-rendered Jinja templates, bespoke CSS design system, light/dark
Database	SQLite (dev) / MySQL (prod) via the same ORM
Deployment	gunicorn + nginx, Let's Encrypt SSL on a Linux VPS

Skills Demonstrated¶

• End-to-end ML web application in a clean, layered architecture
• Training, ensembling and explaining scikit-learn classifiers
• Authentication, persistence and an admin workflow
• Production deployment (gunicorn + nginx + SSL) with a portable SQLite→MySQL data layer

← Back to Projects