Personalizing air pollution exposure estimation using wireless sensor network and machine learning approaches

Abstract

Metropolitan air pollution is a growing concern in both developing and developed countries because of its adverse impact on human health. Most countries have monitoring systems that estimate pollution at the regional level; however, little has been done by way of understanding exposure at the individual level, or to manage such exposure by adapting daily routines. This thesis is an attempt to develop and evaluate a system that crowd-sources air pollution readings taken at fine spatial granularity using mobile sensors, integrating such measurements with fixed-site data to estimate pollution surface using machine learning models, and determining individual exposure by combining pollution surface information with personal location and activity data.

We begin by surveying previous studies on air pollution measurement and health impacts, and argue that personalization is needed for better medical inferencing. Our first contribution is to design and evaluate HazeWatch, a low-cost participatory sensing system that uses a combination of portable mobile sensor units, smartphones, cloud computing, and mobile applications to measure and aggregate fine-grained air pollution concentrations for the Sydney metropolis. The trial results show that our system yields much more accurate exposure estimates than current systems. Our second contribution is HazeEst, a machine-learning-based system that combines sparse fixed-station data with dense mobile sensor data to estimate the air pollution surface for any given hour and day in Sydney, showing that long-term air pollution can be accurately estimated. We show that HazeEst not only yields high spatial resolution estimates that correspond well with the pollution surface obtained from fixed-station and mobile sensor monitoring systems, but can also indicate clear boundaries of the polluted area. Our final contribution is to develop HazeDose, a system that combines air pollution and human activity data to give individuals personalized air pollution exposure estimates, and recommendations on how personal dosage can be reduced. This thesis paves the way towards a deeper understanding of the relationship between air pollution and human health via personalized estimations.