Identification of indoor air quality events using a K-means clustering analysis of gas sensors data

Abstract

Commercial miniature gas sensors, because they are smaller and cheaper than conventional instruments, can be deployed in large numbers to investigate indoor air quality, for research and operational purposes. To compensate for their limited metrological performances, it is necessary to develop relevant data treatment procedures. We applied an unsupervised classification approach based on the bisecting K-means algorithm to data acquired by online gas analyzers and by miniature sensors during a measurement campaign in a low energy school building. This procedure, applied to the analyzers measurements, was able to distinguish the ventilation status and the specific air quality events taking place in the classroom. The same procedure applied to the data from the sensors, even though they were not calibrated beforehand, was also able to identify the same events. The good agreement between the two sets of results validates the methodology and opens up new perspectives for a massive deployment of sensors inside buildings.