We see machine learning discussed everywhere now, but not often in the context of Location. But that would be useful wouldn’t it? We all have data that tells a story about what has happened where in the past, so couldn’t we use that to tell us something about what might happen where in the future? We have seen the beginnings of this in policing, with Predictive Policing but not much beyond that so far.
How does it work?
The problem for location prediction is the same as any other machine learning task. We are looking for patterns in the data and trying to discriminate between events that are causal (i.e. contribute to the pattern) or just random. And sometimes random events lead to others, creating a pattern in the future.
One simulation is only one possible outcome
But prediction isn’t perfect. One simulated future is only one possible outcome. Just like the weather forecast, we look across multiple forecasts to see where there are similar outcomes. This gives us a much better sense of the variability of outcomes. The less variability, the more confidence we can have that the prediction will be accurate.
Can I use it?
So, what does this have to do with me? I might have some data, but I don’t have a supercomputer like the weather forecasters, and I am not a data scientist. Mapcite has location prediction software that might be able to help. Used in conjunction with its other visualisation software tools, you can see the results on a map, so they are easier to understand. You can visualise simulations individually and together. Viewing aggregated predictions is a very simple way to see the commonality.
Perhaps one of the benefits of enforced isolation is that we have the chance to stand back and explore ideas that we might not normally have time to do. If you have some data and want to explore what we can do, let me know.