Air pollution has been one of the major agendas around the globe in recent years. With rising awareness among all citizens, it's of extraordinary importance to measure data related to air pollution in order to have interested parties making informative decisions. Network composed of IoT devices has been one of the tools researchers relied on. Recent years saw rapid progress in sensor technology, which in turn flourished the market for low-cost sensors, giving citizens opportunities to measure various physical properties with affordable and portable sensors. Countless organizations have deployed wireless sensor networks (WSN) involving the usage of IoT devices and budget-friendly sensing hardware. Statistical Analysis of Networks and Systems (SANS) is a research group of the Computer Architecture Department at the Polytechnic University of Catalonia, which has launched several campaigns using WSN composed of Captor devices. Researchers then use relevant machine learning techniques to provide more meaningful information out of otherwise flawed data. A platform using a technology stack composed of Captor devices and machine learning techniques has gone through several stages and is still in the progress of improving itself. This thesis discusses the latest iteration of such a platform, by means of introducing characteristics of hardware, software, as well as machine learning methodologies used. By overviewing and comparing the older iterations of Captor and similar platforms used by other researchers, this thesis hopes to serve as a reference outlook into the current and future development of WSN (with focus on air-quality), where innovations are constantly needed to improve its capabilities. The result of the thesis is an autonomous IoT device, Captor4b, that is self-sufficient for at least 1 month and half where the autonomy can be further tweaked by adjusting duty cycle of Raspberry Pi and Arduino Nano separately from a software perspective.
