erickyegon/immunization-defaulter-risk-engine

Production ML pipeline predicting immunization defaulter risk for 6,864 children across 4,672 CHW areas in Kenya. XGBoost + SHAP + FastAPI + PostgreSQL. Built for Community Health Worker daily prioritisation. # Immunization Defaulter Risk Engine ### Production ML Pipeline · XGBoost + SHAP · FastAPI · PostgreSQL · Kenya CHW Platform > **Try it live:** immunizationengine.streamli… > *Hosted on Streamlit Community Cloud free tier — if the app is sleeping, click "Yes, get this app back up!" and it will resume within ~30 seconds.* --- ## Dashboard Screenshots | 🔐 Administrator View | 👤 User View (CHW Supervisor) | |---|---| | Full technical depth: model performance, drift monitor, live PostgreSQL toggle, all four pages | Stakeholder-appropriate view: programme dashboard and patient risk scorer only — no model internals | | | | *Role separation is enforced at the navigation level — locked pages are not visible to Users, not merely blocked.* --- ## The Problem Kenya's Community Health Worker (CHW) program serves **8.5 million individuals** through 4,600+ CHW areas. Each CHW manages 20–35 under-2 children per catchment area but lacks a systematic way to prioritise which children to visit on any given day. Children who miss vaccines do so silently — there is no alert, no flag, no notification. CHWs currently rely on memory and paper registers. The result: preventable outbreaks, missed booster windows, and inequitable coverage across districts. This engine solves that with a **real-time, explainable risk score** delivered to the CHW's mobile app every morning. --- ## Why This Matters for Managed Care The architecture of this engine maps directly onto the core challenges of Medicare and Medicaid managed care: | This Engine | Managed Care Equivalent | |---|---| | CHW prioritisation list | Member outreach prioritisation for preventive gap closure | | Vaccine defaulter probability | Care gap adherence risk score | | 16.5% positive rate with class imbalance handling | Same challenge in Medicare Stars gap-closure programs | | Per-patient SHAP plain-English drivers | Explainability requirement for HIPAA-compliant clinical decision support | | PSI drift m …