Spatiotemporal statistical machine learning (ST-SML): theory, methods, and applications

Machine learning works brilliantly for narrow questions like “is this a cat?” but fails when policymakers ask messy, real-world questions about how poverty, disease, or crime change across space and time. This fellowship develops new statistical machine-learning methods that can handle those complex, shifting patterns—for example, tracking whether housing quality improves unevenly across a country, or whether satellite images can reveal local progress toward global development goals. The problem is that public-sector practitioners cannot wait for bespoke answers to each new crisis. They need reproducible, well-documented workflows they can apply immediately to urgent policy questions as they arise. This research fills that gap by building theory and tools that turn messy spatiotemporal data into reliable, actionable insights. If successful, the work could transform how governments monitor public health, allocate resources, and evaluate policies in real time. It would let officials see, for instance, whether a disease outbreak is accelerating in one city while slowing in another, or whether poverty reduction programmes are working where they are most needed. The impact is on the quiet infrastructure of public decision-making—the data pipelines and analytical methods that underpin evidence-based policy.