Measuring current and projected mismatch of housing stock with needs: a prospective methodology backed on a Long Short-Term Memory machine learning model[Mesurer l’inadéquation actuelle et projetée du stock de logements avec les besoins en logement : une méthodologie prospective basée sur un modèle d'apprentissage automatique à mémoire à court et long terme]

Matching housing stock to actual needs is a major challenge in metropolitan areas, where demand remains high despite increasingly constrained housing production. Especially in Île-de-France (Paris region), this disparity has led to a shortage of appropriate accommodations for households, while some properties remain unsellable for developers. This study aims to propose a new methodology for measuring current and future mismatch between housing stock and needs. Considering a potential housing demand model that accounts for both the inflow of new households and changes affecting the existing housing stock, we aim to anticipate housing needs through 2030 at a fine spatial scale. As this relies on projected demographic variables and housing stock data, we use a Long Short-Term Memory (LSTM) machine learning model to generate forecasts along with associated uncertainty measures. Then, we propose a dissimilarity index analyzing the current and projected mismatch between housing stock and potential demand which measures the difference between the distribution of household sizes and the distribution of main dwellings by number of rooms, calculated for each housing/household typology. Using publicly available datasets from the National Institute of Statistics and Economic Studies (INSEE) (Annual Housing database, Couple-Family-Household Database, Population Database and Income and Education Database) from 2006 to 2021, we calculated and projected both demographic and construction dynamics, as well as the dissimilarity index, in Île-de-France for 4 888 IRIS units (the finest geographical level) in 2021 and 2030. Our results reflects a structural current and projected misalignment in the housing stock in Île-de-France, at IRIS level, with an over-representation of larger dwellings relative to the actual household composition. In other words, in the majority of Île-de-France territories, the available housing is, on average, larger than the size of households.