Context Aware Real Time Air Quality Prediction Using Machine Learning

Overview

Air pollution contributed to 7.9 million deaths in 2023. In 2025, 156 million people lived in areas with an unhealthy amount of air pollution in the US alone. It’s critical for everyone to be aware of pollution levels to plan day-to-day outdoor activities. Official city air quality meters are limited in number (e.g. only 9 in the entire San Diego city area of 372 sq miles), and air quality changes vastly based on location. This research aimed to find out if air quality can be accurately predicted at any given location using meteorological data, geolocation data, nearest city meter reference data, and deep learning. A custom dataset was created using a hand-held air quality index (AQI) monitor calibrated to the city official meter at Carmel Mountain Ranch in zip code 92128. Particulate matter (PM2.5, PM10) and weather data (wind, temperature, humidity) were measured over 2 months from 9 sites around the city meter. The control variables were the AQI monitor, and the wind data API. Experiments were done with different data collection locations and times, and the deep learning algorithm configurations. It was observed that the PM particle levels differed more than 30% compared to the city meter reference data at distance of 1 to 2 miles from the meter. Wind direction had a low level of correlation with the air quality estimation, and removing it boosted the prediction accuracy by 20%. An Artificial Neural Network model with 10 layers was finalized, with the prediction accuracy of 90%. An Android app was built as a proof of concept.

Competition history

• CSEF 2026 Earth & Environmental Sciences (Track 2) (Junior Division) · Entry J-09-10

Related projects

Closest projects by meaning, across every fair and year in the corpus.

Browse more like this

• All Earth and Environmental Sciences projects
• CSEF Earth and Environmental Sciences projects
• Every CSEF 2026 project