johannesgooth/ds-air-pollution-prediction

A data science project for predicting daily PM2.5 levels across Africa using satellite and weather data. Developed high-performing models and deployed an app for accurate air quality prediction. ## Executive Summary This data science project aims to predict daily PM2.5 particulate matter concentrations across various locations in Africa. The dataset, originally provided for a **Zindi challenge in April 2020**, includes ground sensor, satellite, and weather data. Utilizing this data, we’ve developed multiple models that accurately forecast air quality, which is critical for public health and environmental monitoring. This report details the process from data collection to deploying a predictive application, providing insights and methodologies used throughout the project. Our best model is a **Support Vector Regressor (SVR)**, providing a **Root Mean Square Error (RMSE)** of **22.91** on the test set. This performance surpasses the original winning solution's RMSE of **26.0997**, demonstrating that our model **outperforms the winning benchmark** from the Zindi competition by a significant margin. Additionally, models like **KNeighborsRegressor** and **ElasticNet** have achieved RMSE_test values of **23.34** and **23.59** respectively, further solidifying the robustness of our approach. ## Table of Contents 1. Introduction 2. Project Structure 3. Installation 4. Usage 5. Results and Insights 6. Deployment 7. Acknowledgments 8. License ## Introduction The **ds-air-pollution-prediction** project leverages advanced analytics and machine learning to predict PM2.5 levels, aiding in the development of better environmental policies and health advisories. This project is essential for researchers, environmentalists, and policymakers engaged in air quality management. ## Project Structure This project is organized into several Jupyter notebooks, testing scripts, and a Streamlit application that document each phase of the analytical process: 1. **01_data_collection.ipynb**: Data acquisition from various sources including Zindi and NOAA. 2. **02_data_preparation.ipynb**: Data cleaning and preprocessing for analysis readiness. 3. **03_exploratory_data_analysis.i …