Unsupervised K-means learning applied to the investigation of indoor air quality events with electronic gas sensors networks

Indoor environments present high specific pollutant concentrations, especially volatile organic compounds, inducing a risk for human health. Tools able to provide real-time information on indoor air quality (IAQ) are needed, in order to feed into the knowledge of these environments, identify the sources and automatically control remediation systems. Electronic gas micro-sensors are low-cost and nonintrusive instruments, which can be permanently deployed in large numbers to map the air quality of a whole building, but that still need to be evaluated. Five multi-sensors systems, combining several selective and non-selective electronic gas micro-sensors, have been installed inside a classroom at a low energy junior high school, in parallel with a set of standard analytical tools. The aim of this work was to determine if multi-sensors systems, associated with unsupervised k-means learning to process the data, were able to monitor effectively the dynamics of indoor air in a room in real conditions, and at what level of data quality.