One analysis that can be conducted in this context is to see how case-rate changes on applying isolation policies. If a certain POI is closed, populations from each ZCT are redirected to other POIs with similar categories (For example - if grocery store A is closed, then populations tend to visit other nearby grocery stores). This could result in a change in case-rate and weight distribution dynamics. Similarly, we can use our prediction values to see how each POI behaves on restricting population flow from ZCTs with higher case-rates. Such analyses are made possible by our model.

Our method can be extended to use a mechanistic model like here to train a epidemiologically informed model where parameters such as infection rate, recovery rate are interpretable. We can also use more sophisticated methods to learn and leverage relations between POIs and ZCTs such as use Graph Neural Networks. This would enable us to directly leverage sparse network structures for efficient training and provide explainable predictions.