AstridMarion/Air-Quality

Air quality analysis in African cities. # Air Quality Prediction for African Cities: A Data-Driven Public Health Solution > Predicting PM2.5 levels 24-48 hours in advance across four African cities to enable proactive public health interventions. --- ## Table of Contents * Executive Summary * Business Context & Problem Statement * Dataset & Data Understanding * Analytical Approach * Technical Implementation * Results & Model Performance * Geographic Performance Analysis * Business Impact & Recommendations * MLOps Implementation * Limitations & Future Work * Installation & Usage * Project Structure * Authors --- ## Executive Summary Air pollution is responsible for over **7 million premature deaths annually** worldwide, with sub-Saharan Africa experiencing some of the fastest increases in pollution levels due to rapid urbanization. This project delivers a machine learning pipeline capable of predicting PM2.5 concentrations across four African cities (Lagos, Nairobi, Kampala, Bujumbura) with **17.3 µg/m³ average error**, enabling authorities to issue early warnings 24-48 hours before critical pollution episodes. **Key Achievements:** - 53.6% variance explained (R²) using XGBoost, significantly outperforming linear baselines - Geographic generalization validated through strict cross-city validation preventing data leakage - Actionable accuracy: prediction errors within WHO air quality guideline thresholds - Production-ready pipeline with full MLflow experiment tracking and model registry **Business Value:** Early pollution forecasts enable targeted public health interventions (school closures, traffic restrictions, vulnerable population alerts) that can reduce emergency room visits by an estimated 15-20% during peak pollution events. --- ## Business Context & Problem Statement ### The Public Health Crisis Particulate Matter 2.5 (PM2.5) — fine particles smaller than 2.5 micrometers — penetrates deep into lungs and bloodstreams, causing cardiovascular diseases, respiratory infections, and p …