Re-defining Spatial Typologies of Humanitarian Housing Plans Using Machine Learning

The complexity of post-disaster recovery processes and the risks of implementing “one size fits all” solutions are now widely acknowledged by humanitarian actors, who are pushing for innovative approaches to better link housing recovery and urban development after urban disasters. However, there is currently little technical support for responsible local administrations tasked with selecting Temporary Housing (TH) sites, to measure and visualise the complexity of their spatial configuration, illustrating patterns of urbanisation and transportation. The lack of time and appropriate decision-making support systems contribute to perpetuate unsuccessful practices of housing assistance provision, which detract resources from more important long-term investments. Therefore, this paper proposes a computational analysis method which seeks to support the delivery of truly people-centred and performance-based TH plans by enhancing administrations’ capacity to audit TH proposals. It seeks to disclose non-trivial differences and trade-offs in candidate TH plans, by exploiting unsupervised machine learning techniques to quantitatively define, starting from existing data, similar types of spatial arrangements. This research assumes that clustering algorithms can establish deep analogies between diachronic urban configurations, both real and projected. The proposed computational method can be scaled up so as to handle multiple dimensions according to changing strategic planning priorities and varying local needs.