According to the UN Environmental Program, air pollution is the greatest threat to public health and the environment globally and accounts for an estimated 7 million premature deaths every year. Previous prediction modeling has been used to detect same-day air quality using the current day’s data, but nothing has ventured into the future days. This is essential to local communities to warn citizens against potential risks due to dangerous levels.  

 In this project, an Artificial Neural Network (ANN) model was developed to predict air quality using multiple weather and climate-based variables. One particular focus of this project was to predict air quality as many days into the future as possible.

 The datasets used in this project contain measurements of different weather parameters and a corresponding air quality level from 2011 to 2020 for four different US counties with air quality issues. Weather variables included in the dataset are temperature, humidity, precipitation, wind speed, and direction. Hyperparameter tuning was done for the ANN models to optimize prediction capabilities. 

The results from training on separate databases showed that the models were able to achieve an accuracy in the range of 75% to 96%, which is considered a successful prediction. Hyperparameter tuning on the daily data samples displayed minimal variation in the testing accuracy. Model exploration to predict air quality for a future day displayed that the accuracy was maintained for up to 3 days into the future. This modeling potentially could be implemented as an advanced warning system for city governments. 
 

​

[Git Repository]