Infectious diseases used to spread from city to neighbouring city. But as the world grew more connected, this changed. Diseases now spread globally by long-distance travel.
Epidemiologists have tried to model this using complex algorithms and powerful computers. It hasn’t been easy. And the results look almost random, as seen in this video:
But instead of using complex models, you can shift your perspective. Today, geography isn’t most important for the spread of diseases. What’s most important is the frequency of air travel between airports.
You can map the frequency of flights by route. You can then map how diseases spread through routes. This has more explanatory power, as you can see here:
I learned this today in a presentation by complex systems researcher Dirk Brockmann. It was a good reminder that with the right perspective, simple models can often beat complex. (_HBR_ has more on this this here and here.)
As I work in machine learning, I’m surrounded by models with many features and parameters. But more features and more parameters doesn’t always mean more accurate.