A stepwise approach integrating feature selection, regression techniques and cluster analysis to identify primary retrofit interventions on large stocks of buildings

•This research presents a simplified tool to analyse the energy performance of the building stock.•The method combines Wrapper Feature Selection, Random Forests, Hierarchical and k-medoids clustering.•Buildings are grouped according to their main features and reference buildings are identified.•Energy efficiency measures are prioritized for the different groups.•The proposed method is applied to a set of 41 Italian schools. In the recent years, existing public buildings have been put under the spotlight for the application of retrofit strategies prescribed by the European Energy Efficiency Directives. Among them, schools have a pivotal role since, besides energy performance, they have to cope also with high indoor environmental quality requirements. However, the definition of a refurbishment policy for the stock of school buildings presents some criticalities: limited data are often available, comprehensive energy audits are too onerous to apply to each school building, and the findings of many case-studies discussed in the literature can be too specific for a robust generalization. In this context, this work proposes a new integrated method for energy audit on large stocks of existing buildings, avoiding case-by-case analyses and focusing on identifying the most significant retrofit areas and priorities of intervention. This approach, based on the combination of different data mining techniques (i.e., Wrapper Feature Selection, Random Forests, Hierarchical and k-medoids Clustering), is meant to deliver a useful tool for the existing buildings’ stock to professionals and Public Administrations. The method is described and discussed, and then applied for validation purpose on a case study of 41 educational buildings in the Province of Treviso, Italy.