Anita Ruangrotsakun
This project is a much simpler, smaller version of of the work Caleb Robinson and Bistra Dilkina completed in "A Machine Learning Approach to Modeling Human Migration". They developed a machine learning framework to improve predictions for human migration patterns and compared it to several traditional migration models. Some of their implementation code can be found here: Caleb Robinson's Python library for working with migration data.
The scope of this project was constrained to implementing only a radiation model and gravity model in R, as opposed to Python. I also make use of the historical migration data from the IRS (as referenced in the article), though I use the in and out flow files for states instead of US counties, to make predictions about migration patterns within the US in the next few years.
A more complete project report can be found /documentation/project_report. Notes I took while doing background research can be found in /documentation/project_notes
The raw data can be found in /data.The migration data files are grouped by year, whereas the population estimates for 2009-2019 can be found under /ACStables.
In /R, you will find three files containing the functions I wrote as well as their documentation. Some unit tests for the most important functions can be found in /tests/testthat.
The /analysis folder contains the scripts I created to:
- generate migration matrices using the gravity and radiation models
- clean up the historical data (put them all in a standard format)
- compare the historical data to the models' predictions
- fine tune the models and use them to simulate future migration flows (2018-2022)
If you do not already have these packages, you may need to run:
install.packages("Rmpfr")install.packages("Imap")