Perry-Bradley/cmr-malnutrition-

Cameroon Malnutrition statistics # Cameroon Malnutrition Atlas **Course:** CEC 420 — Data Mining **Author:** SEPO PERRY-BRADLEY DINGA (CT23A145) **Department:** Computer Engineering — Software Engineering, University of Buea **Academic Year:** 2025 / 2026 A CRISP-DM project that identifies the strongest drivers of child stunting in Cameroon and produces a ranked list of high-risk regions using **real DHS & MICS sub-national data** from 1991–2018 (the only public sub-national survey series Cameroon has released). Four data-mining techniques are applied: | Technique | What it answers | |-----------------|-----------------------------------------------------------------------------------| | **Regression** | What's the predicted stunting % for each region? | | **Classification** | Which WHO risk band (low / medium / high / critical) does each region fall in?| | **Clustering** | Which regions share a similar driver profile? | | **Forecasting** | Where will each region be in 2026 / 2028 if its trend continues? | Plus an **in-browser predictor** (the `/predict` page) — exported linear, logistic and K-Means models that run client-side with no backend. ## Real data, not synthetic The pipeline uses **only real Cameroon DHS / MICS values**: - Sub-national stunting + 13 driver features for **10 regions × 5 DHS rounds = 50 rows**. - Survey years: **1991, 1998, 2004, 2011, 2018**. - 1991/1998 only published five mega-regions (e.g. "Adamaoua/Nord/Extrême-Nord"); we broadcast each mega value to its constituent modern regions so the time series is complete. - No newer sub-national survey for Cameroon exists in the public domain yet; forecasts to 2026/2028 are linear extrapolations from this real series. There is no synthetic-data fallback any more. ## Quick start ```powershell # 1. Install Python dependencies (Python 3.11) & "C:\Users\USER\AppData\ …