Classification of dwellings into profiles regarding indoor air quality, and identification of indoor air pollution determinant factors

This study aims to identify the most relevant variables, among outdoor measures, building characteristics and socioeconomic situation, for predicting indoor air chemical pollution in dwellings. To achieve this, we propose a two-step plan: first, group the dwellings into classes according to the indoor measured concentrations, then use regression tools to express a dwelling’s class as a function of the aforementioned variables. In the first step, we use modelbased clustering algorithms in a multivariate Gaussian mixture context; in the second step, we use binary decision trees in a discrimination context. This approach does not handle the pollutants individually, but considers them together as a multidimensional variable that must be summarized by a categorical variable (the dwelling’s class).