Using Neural Networks to Predict Air Pollution Based on Traffic Patterns and Weather Data

Airborne pollution due to the use of internal combustion engine vehicles can have a detrimental impact on human health and the environment. Criteria airborne pollutants, including particulate matter (PM) and sulfur dioxide (SO₂) can cause serious illnesses and environmental damage including hormonal and behavioral problems, respiratory illnesses, acid rain, smog, aggravated asthma, decreased lung functions, heart attacks and death, among others. Impoverished areas within Alabama can easily be affected by increased air pollution with no strong government action to help alleviate the situation. Prichard, AL is a perfect example of such an area. Typically Prichard does not experience heavy external traffic, however, trailer trucks and buses can be redirected from the nearby tunnel to pass through Prichard under dangerous weather conditions. This increased traffic is occasional, and should have a limited effect on the air pollution levels in the area. This project focuses on developing an artificial intelligence mechanism that will predict the resultant air pollution from given traffic and weather data. After analysis of different types of AI that could be used, our team decided to develop a neural network that will be able to make accurate air pollution predictions by analyzing previous traffic data from surrounding highways, meteorological data and resultant air pollution data. Neural networks are structured like a brain. They are built from layers of “neurons” that receive an input and produce an output using a series of algorithms. If provided enough information, a neural network is able to recognize underlying patterns and trends within the data. Therefore, when provided with new inputs, the neural network is able to produce an output that serves as a good approximation based on previously learned information. In this project, more specifically, we plan to build a neural network that is able to provide accurate predictions of the resultant PM and SO₂ air pollution based on changes in traffic patterns. For a neural network to produce accurate results it is important to also take into account all environmental factors that can affect the output variable, for example, air temperature, air humidity, wind speeds, and precipitation. After collecting daily data from each of these data types, we plan to begin building the neural network. This research will help determine the factors that contribute the most to air quality in Prichard, while also creating a predictive model for air pollution based on future traffic patterns in the area.