aduku-titus/ncd-stockout-predictor

A machine learning project to predict stock-outs of essential NCD medications in Ghana. # NCD Medication Stock-out Predictor *A proof-of-concept project exploring the use of interpretable machine learning to support health system resilience in resource-constrained settings.* --- ## 1. Clinical Motivation & Context As a Registered Nurse with over five years of experience in a resource-constrained district hospital in Ghana, I have repeatedly witnessed the direct, negative impact of medication stock-outs for chronic Non-Communicable Diseases (NCDs). These events disrupt the continuity of care for vulnerable patients, erode patient trust, and place an immense administrative burden on clinical staff. This project is a technical probe designed to move from frontline observation to proactive intervention. It explores a fundamental question: **Can we use historical data to build a simple, reliable, and interpretable early warning system to help pharmacists and nurses anticipate and mitigate these critical supply chain failures?** ## 2. Project Overview & Key Features This repository contains the end-to-end development of a prototype to predict stock-outs for essential NCD medications. * **Data Engineering:** A realistic, messy dataset was synthetically engineered to mirror the data quality and feature types typically found in a district-level Health Information Management System (HIMS) in Ghana. * **Model Development:** A Logistic Regression model was trained using `Scikit-learn` to provide a robust and easily understandable baseline for prediction. * **Model Interpretability (Crucial for Trust):** The `SHAP` (SHapley Additive exPlanations) library was used to explain every individual prediction. This is a critical feature, as clinical adoption of AI tools is impossible without transparency and trust. We must be able to answer *why* a model is making a certain prediction. * **Interactive Dashboard:** A human-centered, interactive dashboard was built using `Streamlit` to allow non-technical clinical staff to input data and receive a clear pre …