README.md

metadata

title: PM2.5 Air Quality Predictor
emoji: 🌍
colorFrom: blue
colorTo: green
sdk: docker
pinned: false
app_port: 7860

Air Quality Prediction with Explainable AI

A PM2.5 air quality prediction system built with XGBoost, featuring explainability through SHAP and counterfactual analysis.

Overview

This project predicts daily PM2.5 concentrations using historical air quality measurements from the OpenAQ dataset. The model is accompanied by explainability tools to help understand predictions and explore what-if scenarios.

Features

• XGBoost-based PM2.5 prediction model
• SHAP analysis for feature importance and local explanations
• DiCE counterfactual generation for scenario analysis
• Interactive Streamlit dashboard for predictions and visualizations
• Global air quality station coverage using OpenAQ data

Quick Start

Simply use the app above to:

1. Search for air quality stations globally or geocode any location
2. Select a date to get PM2.5 predictions
3. View SHAP explanations showing which historical factors influenced the prediction
4. Explore counterfactual scenarios to understand what changes would improve air quality

Data Source

This project uses the OpenAQ Open Data on AWS archive:

• S3 bucket: s3://openaq-data-archive/
• HTTP: https://openaq-data-archive.s3.amazonaws.com/

No API key required - data is accessed directly from the public archive.

Technical Details

• Model: XGBoost regression
• Features: Historical PM2.5 lag features (1-day, 7-day, rolling averages)
• XAI Methods: SHAP for local explanations, DiCE for counterfactuals
• Data: OpenAQ global air quality measurements