A classifier ensemble for classification of dynamic data. Application to an indoor air quality problem

Indoor air quality has an important impact on people exposure to pollutants. The Airbox Lab company currently designs a connected object, called Footbot, measuring every minute several different parameters related to indoor air quality : temperature, humidity, VOC concentrations, CO2, formaldehyde and particle matter (pm). Moreover, Footbot ought to include some data analysis features to identify different domestic situations (presence, cooking, housework and so on) from the gathered data. The final purpose is to help user avoiding situations causing air quality degradation. In this paper, two different tools (neural networks and decision trees) are tested and compared to solve this problem of dynamic data classification. To increase the classifier performances, classifier ensembles are also studied. © Lavoisier 2016.