Global Migration and CARTs

Measuring Migration - Why Does it Matter?

…

Understanding the link between climate change and migration

Source: http://www.internal-displacement.org/globalreport2016/

Migration - The Big Picture

…

We are interested in a world-wide comprehensive persepctive

Source: Abel, G. J., & Sander, N. (2014). Quantifying global international migration flows. Science, 343(6178), 1520-1522.

Migration - The Big Picture

…

In order to get there, we need data

The Bigger the Better

…

IPUMS gives us large data about internal and external migration...

• covering a time interval from 1962 to 2011,
• including 58 countries and 164 "country-years",
• with 78mn (originally >400mn) observations.
• Our features include sex, age, occupation, education, emplyoment, year of survey, and continent.

The Bigger the Better

…

for us and for Classification Algortihms

Source: https://en.wikipedia.org/wiki/Decision_tree_learning

Classification Trees

…

split the parameter space in the most "efficient" way

Source: James, G., Witten, D., Hastie, T., & Tibshirani, R. (2013). An introduction to statistical learning (Vol. 112). New York: springer.

Classification Trees

…

split the parameter space in the most "efficient" way

Source: James, G., Witten, D., Hastie, T., & Tibshirani, R. (2013). An introduction to statistical learning (Vol. 112). New York: springer.

Classification Trees

…

split the parameter space in the most "efficient" way

Source: James, G., Witten, D., Hastie, T., & Tibshirani, R. (2013). An introduction to statistical learning (Vol. 112). New York: springer.

Our First Classification Tree

…

classifies no (0), internal (1), and external (2) migration of 78mn observations

Taking a Closer Look

…

we can see that education and age matter

Next Steps

…

for which your feedback is very appreciated

• Are there more "effective" ways to perform classification (with less trade-off in sample size)?
• How can we bring climate change on a macro-level effectively into the picture?
• Could machine learning techniques help us to auto-complete our missing data?