Cross-dwelling validation of indoor environmental monitoring for operational risk screening in social housing portfolios

Abstract

Social housing providers use indoor environmental monitoring within asset management systems. The extent to which these data can differentiate operational risk domains across independent dwellings has not been fully evaluated in operational deployment. Current predictive modelling frequently relies on random data partitioning, failing to reflect situations in which models are applied to previously unseen properties. This study examines the cross-dwelling explanatory capacity of environmental exposure indicators within a London-based
housing portfolio. Five years of monitoring data from 93 UK social housing dwellings were linked with operational risk records, yielding 5,748 monthly dwelling-level observations. Indicators derived from temperature, relative humidity, and carbon dioxide were analysed using Ridge regression and Random Forest models under five-fold property-grouped cross-validation. Under grouped validation, indoor air quality and excess heat domains show positive explanatory power across dwellings. In contrast, envelope-related domains, including heat loss, draught, and cold home risks, produce near-zero or negative R² values, indicating limited cross-dwelling information in bulk indoor environmental measurements. Random Forest models do not consistently improve over regularised linear models. These findings identify the risk domains that can be informed by environmental screening at portfolio level and those which require further or direct structural assessment within asset management practice.