Deep Learning Inference of Personalized Air Quality Index

Air pollution is the world’s largest single environmental health risk according the the World Health Organization. Data about the pollution is publicly available in almost any location. Companies such as Google and Apple have started showing the pollution information in the weather app of Android and iOS devices. The pollution information that they provide is usually the concentration level of a certain pollutant, or an index that makes easier for people to understand the risk level. The problem with this method is that it gives the risk level considering just the pollutant(s) concentration but does not involve any information about the person. If a person is resting at a certain location, the risk of being exposed to a pollutant is lower than a situation in which the person is running at the same location. If the person is running, they are inhaling more air, thus more pollution, than when they are resting. In this thesis we propose a methodology to personalize the risk of being exposed to a certain pollutant. We talk about a possible way to estimate the PM2.5 concentration at any location in Chicago. We discuss a methodology to estimate the heart rate of the user. This work also contains methods to compute, estimate, and personalize the minute ventilation; it illustrates how the values estimated in the previous chapters are used to show to the user the personalized risk. We also talk about how we built the MY-AIR app and the data that we have used.