darknuma/databricks-team

datafest africa # **⚕️ Pre-Emptive PHC: AI-Powered Supply Chain Forecasting** **A predictive analytics dashboard to prevent stock-outs of essential medicines in Nigeria's Primary Healthcare Centers (PHCs).** ## **🚀 The Challenge: The Last Mile Problem** Inspired by the Nigeria Health Supply Chain Program Challenge, our team tackled a critical issue: **how can we anticipate demand for essential medicines when direct consumption data is unavailable?** Stock-outs at the local level threaten public health, and traditional supply methods are often reactive, not proactive. ## **💡 Our Solution: Proxy-Based Demand Forecasting** We built a system that predicts future demand for antimalarial drugs (ACTs) at the Local Government Area (LGA) level. Instead of relying on non-existent stock data, we engineered a **"proxy demand"** variable by analyzing household survey data (DHS/MIS) for indicators like fever prevalence and treatment-seeking behavior. This data, enriched with geospatial information on health facility density, powers a machine learning model that generates a 3-month rolling forecast, visualized on an interactive risk map. We wrote about the EVIDENCE in `Evidence PHC.md` and you will find the analysis in `analysis.ipynb` ### **✨ Live Demo** *check demo* ## **🛠️ Tech Stack & Architecture** We chose a modern, scalable stack to build a robust and performant solution. | Component | Technology | Description | | ----- | ----- | ----- | | **Backend API** | \ **FastAPI** | For a high-performance, auto-documenting API to serve model predictions. | | **ML/Data** | \ **Pandas/GeoPandas**, **Scikit-learn**, **Joblib** | For data wrangling, geospatial feature engineering, model training, and serialization. | | **Frontend** | \ **Tailwind CSS**, \ **Leaflet.js** | For a responsive UI and a rich, interactive choropleth map visualization. | | **Data Sources** | **DHS/MIS, NMIS, OpenStreetMap** | Fusing national surveys with open geospatial data on health facility locations. | ### …