Home » Research » Using Machine Learning to build an Asylum Coalition Index

Using Machine Learning to build an Asylum Coalition Index

  • The Asylum Coalition Index (ACI) uses machine learning to group European governments according to shared priorities on irregular migration and asylum. By identifying where two or more European countries share an origin country of concern, governments have the information they need to build strategic partnerships.
  • The ACI is created using an automated tool called “k-means clustering.” This tool assigns scores to European countries based on the proportion of its asylum seekers that come from each origin country. It then groups European countries that share high scoring origin countries.
  • Results suggest that in 2020, Austria would be well-served by working on origin country asylum issues with Greece, Hungary and Belgium. Denmark’s best return on investment would likely be with Sweden, the Netherlands and Norway. For France, the best places to start would be the UK, Italy and Switzerland.

On asylum, cooperation is key

When observing political alignments and shuffling coalitions on migration and asylum policy in Europe, bold announcements, angry dissent and persuasive (or coercive) diplomatic dances are behind the most engaging action.

For national governments, the domestic politics of migration can become so urgent that they become a primary influence over approaches to coalition-building at the European level. This can lead governments into costly trade-offs, for example, taking a stand over EU relocation at the cost of losing influence over budget decisions.

These coalition decisions are fundamentally political. However, there are some objective facts that influence how beneficial or costly the decisions may be.

European government have specific priorities

In migration and asylum debates, different countries have different asylum-seeker demographics. One country may attract more Syrians, another more Eritreans. One may attract young men, while another may attract more unaccompanied minors.

These differences can be important to that country’s domestic policy choices and foreign policy engagements. If Sweden receives high numbers of asylum applications from Afghan nationals but Denmark has a high influx from Nigeria, their priorities might not be best served working together – even if they otherwise cooperate well.

By the same token, similarities can form the basis of alignment and make cooperation feel natural. If Germany is facing high numbers of spontaneous arrivals from Nigeria and the United Kingdom is too, then this shared priority might form the basis for common action.

Naturally, at any given moment, foreign policy coalitions may be influenced by more powerful drivers than asylum. Perhaps both countries have a longstanding set of investments in the country of origin, or a history of political mediation. Nevertheless, a country may generate more benefit for less effort if its foreign engagements to influence asylum-seeking trends are built around the most likely coalition partners.

Making the Asylum Coalition Index

This Asylum Coalition Index (ACI) is a method-under-development for helping policymakers identify which partnerships make most sense.

In each year, for each European country, we tallied the total number of asylum-seekers in that country, then calculated the proportion of asylum-seekers from each country of origin. Each European and origin country pair was given a score based on these criteria. If Spain reported 10,000 asylum-seekers in 2014 and 1,500 of these were from Eritrea, then the Spain-Eritrea pair for that year received a score of 0.15. If Spain had zero asylum-seekers from Somalia, then the Spain-Somalia pair for that year received a score of 0.0.

What is k-means clustering?

k-means clustering is one of the machine learning tools that data scientists use to quickly make sense of unstructured data. It uses an algorithm that decides the most relevant ways to put data into clusters or groups, based on the criteria we care about.

This means that we don’t need a human to hypothesise or guess which groups might belong together but can ask the algorithm to work out which groups make most sense based on the data itself.

Examples of use include: a business finding ways to categorise people according to their purchasing behaviour or a healthcare provider grouping diabetes patients according to their underlying conditions.

We then applied the k-means clustering algorithm to the dataset. The algorithm tries to group European states based on how closely they share proportions of asylum-seekers by country of origin. We told the algorithm to sort all European states into four groups.

A group contains countries that have the closest shared interests in terms of their foreign policy engagements around asylum-seeker issues. This works on priorities: a country in group one might share interests with countries in other groups, but the relationship is weaker than countries in its own group.

“Natural” coalitions on asylum-seeker policy’s

The maps below show groups in three years: 2008, 2013, 2018.

Between 2008 and 2013, we see shifts in coalition groups reflecting changing trends in the desire and practical possibility of asylum seekers to reach certain destinations. To some extent, these changes may reflect the result of poor coalition building in the past, where unilateral reactions to irregular arrivals displaced problems to other countries that, in fact, shared the same strategic objectives. While a Europe-wide agreement on how to respond to transcontinental inflows is unlikely, strategic coalitions might reap dividends.

Predicting your best partnerships

For any government official, this approach offers an efficient opportunity to analyse potential coalitions. There are many applications, depending on what you are trying to achieve through coalition-building.

For example, imagine you work for the Austrian Government. You run this analysis to give you coalitions year by year. However, you know that European coalition-building and foreign engagements take time, so you decide that you will combine recent years to get a sense of your most stable coalition partners. For each European country, you check how many times you have been in the same group between 2008 and 2018. You also decide that more recent years are more relevant, so you give one point when Austria shared a group with others in 2018 but only half a point when they did in 2013.

Using this example approach, the tables below suggest that Austria in 2020 would be well-served by sitting down with Greece, Hungary and Belgium to compare asylum-seeker demographic patterns, then considering what this means for coordinated action in Brussels or with countries of origin. By contrast, Denmark’s best return on investment in coalition discussions would likely be with Sweden, the Netherlands and Norway. And for France, the best places to start would be the UK, Italy and Switzerland. The benefit of this approach is that it quickly generates operational hints to coalition-building based on sustained interests and facts on the ground.

K-means clustering best partnership

From Demonstration to Deployment

This demonstration has used just a single type of data. Seefar’s policy analysis team has developed the approach more fully by combining it with additional data streams, such as foreign resident data beyond asylum seekers and other measures of foreign policy engagement with the country of origin.

We are also exploring other applications:

  • Country of origin focus – inverting the analysis lets us look at the most important countries of destination from the perspective of the country of origin. This would allow foreign ministries and embassies to identify the strongest European coalition for engagement with that country.
  • Future intentions – using Seefar and partner databases containing information on the migration plans of hundreds of thousands of potential migrants in Africa and Asia, we can infer how trends may change in the next year.

Sign up for updates or get in touch with us directly here.