toluwanimisamuel24/Malaria-Incidence-Trends--Nigeria--2010-2023

State-level malaria incidence analysis for Nigeria (2010–2023) using CRISP-DM. Combines Malaria Atlas Project and NASA POWER climate data, a validated Random Forest model with SHAP explainability, and an interactive Power BI dashboard. Built for the 3MTT data analytics programme. # Malaria Incidence Trends in Nigeria ### A CRISP-DM Data Analysis, Machine Learning & Power BI Dashboard Project **Author:** Raji Toluwanimi Samuel **Track:** AnalystLab Africa Data Analytics and Machine Learning Internship Programme **Methodology:** CRISP-DM (Cross-Industry Standard Process for Data Mining) --- ## 1. Business Understanding **Question:** What are the trends in malaria incidence across Nigerian states from 2010–2023, and how much can climate data and historical incidence explain about a state's future burden? **Why it matters:** Nigeria carries one of the world's highest malaria burdens. Resource allocation decisions where to prioritize nets, chemoprevention campaigns, and surveillance benefit from understanding both *where* burden is concentrated and *how predictable* it is from year to year. This project builds a state-level analytical pipeline to support that kind of prioritization, and is transparent about what the data can and can't explain. **Intended audience:** A public-health decision-maker (e.g. NMEP-style program officer) who needs a clear, honest picture of state-level trends — not a black-box prediction. --- ## 2. Data Understanding | Source | What it provides | Level | Coverage | |---|---|---|---| | Malaria Atlas Project | Malaria incidence per 1,000 population | State (admin1) | Nigeria, 2000–2024 | | NASA POWER API | Rainfall & temperature | State (via centroid lat/lon) | Nigeria, 2010–2023, monthly | Both sources are public and required no authentication. The overlapping usable window across both sources is **2010–2023**, which defines the project's analytical scope. **A data-quality issue found and fixed during this project:** NASA POWER's monthly rainfall parameter (`PRECTOTCORR`) returns an *average daily rate* (mm/day) for each month, not a monthly total. An earlier version of the pipeline summed these 12 raw values directly, producing annual rainfall figures roughly 30x too low (e.g. ~40mm/year for FCT, when the true …