The global refugee crisis has been going on for quite a while now. According to the United Nations, an incredible 65 million people have been displaced from their home globally. Per day, more than 28,000 people are forced to flee their surroundings.
A recently published study aims to streamline and answer the pressing question – where should the authorities place the refugees to optimize their skills? Currently, the process is done manually and in a lot of cases, at random. This leads to a mismatch between what refugees can offer, and where they are actually placed and what they’re asked to do.
A team from Stanford University has developed a machine learning algorithm for geographically placing refugees to maximize their overall employment rate. The algorithm uses features like country of origin, gender, languages, and age to train the model. The training set for the algorithm had 33,000 working age refugees in the period between 2011 and 2016. The model was then tested on 900 refugees entering the United States and Switzerland at the end of 2016.
The results of the model were pretty impressive. According to the report, “The median refugee’s predicted probability of employment in the United States more than doubled, increasing from approximately 25% to 50%”. 34% of the refugees in the test group found work within three months of arriving in the United States. Using this algorithm (and fine-tuning it more) the researchers predict this figure could climb up to 48%.
Figures from the United States model on the test set
In the case of Switzerland, the model found that the ability to speak French resulted in a much better pay-off for refugees assigned to French-speaking cantons, rather than German speaking cantons.
Figures from the Switzerland model on the test set
You can read the full Stanford research report here.
Our take on this
This is a wonderful example of data and machine learning being put to a great use in our society. The refugee crisis is a very real thing and hopefully this research will help them settle down and stabilize themselves. The algorithm still needs to be worked upon to improve the accuracy – test it on more countries, use a bigger dataset, test for longer and shorter term results, etc.
If the authorities start using their current processes with the machine learning model, it will lead to a better overall experience for everyone involved.