juliettefaurie/AirQualityDataAnalysis

Production-ready ML pipeline to predict air quality (PM₂.₅) in Africa using satellite data, featuring geographic cross-validation and MLflow tracking. # Production-Ready ML Pipeline: African Air Quality Prediction (PM₂.₅) A machine learning pipeline for air quality prediction using time-series data with geographic features, aligned with the **Evaluation** phase of the **CRISP-DM** framework. *Authors: FAURIE Juliette, PHAM Ngoc Thu Uyen and POKHAREL Sushant* ## 1. Context & Business Understanding ### Overview and Goal The goal of this project is to predict PM₂.₅ air pollution levels for cities across Africa. The pipeline leverages authentic satellite measurements of gaseous pollutants, cloud parameters, temporal patterns, and geospatial features. The technical implementation utilizes standard libraries (Pandas, NumPy, Scikit-learn) and advanced models: **XGBoost** and **LightGBM**. ### Business Objective The primary focus is to transform raw environmental data into **actionable insights** suitable for public health policy. 1. **Technical Objective**: Develop robust, non-linear models capable of estimating PM₂.₅ concentration (µg/m³) with minimal error. 2. **Business Objective**: Provide reliable short-term forecasts to municipal authorities and health agencies to enable **timely public advisories** and inform targeted mitigation strategies. ## 2. Dataset Description ### Data Coverage * **Data Source**: Satellite atmospheric data collected from multiple locations across Africa. * **Period**: 01/01/2023 - 26/02/2024. * **Training Set**: 8,071 records, 80 features. * **Test Set**: 2,783 records, 79 features (PM₂.₅ is the missing target to predict). ### Key Variables | Feature Type | Variables | Description | | -------------- | -------------- | ------------------ | | **Target Variable** | PM₂.₅ | PM₂.₅ concentration (µg/m³). | | **Temporal Features** | date, hour | Capture seasonal, weekly, and daily pollution cycles. | | **Geographical Features** | site_latitude, site_longitude, site_id, country, city | Help the model recognize pollution characteristics by geographic location. | | **Atmospheric Features** …