Project Jetson is a predictive analytics project aimed at providing predictions on displaced population movement for the Somalia situation launched by UNHCR’s Innovation Service. Jetson is focused on predicting forced displacement movement, through analyzing external factors and variables as well as internal UNHCR data. The purpose of this website is to enhance information-sharing of data innovation processes and the results of the experiment. The information shown on this site should be viewed as indicators of potential movements and their underlying causes. Therefore, interpretation and application of predictions data is at your own discretion. If you would like to know more about Project Jetson, please reach out to UNHCR Innovation Service, at innovation@unhcr.org.

One of the most pressing challenges for humanitarian agencies in trying to predict population movement (population flow) is the information gap that exists when many people move from one place to another. This is particularly challenging during emergencies when the ‘fluid or rapid’ movement makes it difficult for humanitarian actors to track and estimate arrivals, and departures. In consequence, this challenge affects planning, emergency preparedness, funding of assistance, and operational services. The challenge is exacerbated by: A) Lack of access to territory: referring to regions with ongoing violent conflict, to occupied territories and/or geographical areas where humanitarian actors are not present or access is denied. B) Lack of resources: lack of financial or human resources to deploy enumerators to collect data from specific sources or geographical locations. C) Lack of informants: collecting data from specific sources or a particular group can bias the data D) Lack of validation: sometimes information verification is conducted only by participant observation (visual) and/or the data provided by the informants could be partial or inaccurate for diverse number of reasons.

1. To provide predictions on population movement during displacement situations for UNHCR and other partners to make better operational decisions
2. The project also intends to put a spotlight on humanitarian, and development actors to share their data in a systematic way, in order for computers to read it and process it automatically, which means passing from a predictive analytics phase to a prescriptive analytics phase.
3. To serve a case of the power of open data. For other developers, statisticians, data and computer scientists to build their own models and applications with UNHCR data.

On forecasting statistical techniques and open-source machine-learning software for modeling. The scripts and the aggregated and anonymized datasets will be published in UNHCR github once the experiment is finished.

We are currently at testing phase, and will eventually start adding more regions and situations to the experiment.

• The project is focused only in the population flow, both across borders (refugees) and internally (IDPs). This means the main research purpose is to measure changes in the population in an specific period of time (e.g. a month).
• There is not one single major cause of forced displacement. There are many causes (variables) and an infinite combination of these variables, along with behavioral human patterns that lead to forced displacement. The objective of the model is not to find the cause (why?) but to explain how are people moving (patterns).
• This project attempts to find interesting and meaningful patterns in data by making assumptions from the data our partners are providing, particularly when data is not available, or is incomplete.
• Some of the assumptions do not need a model to tell that the assumption is correct. Building a model goes beyond identifying the assumptions’ characteristics (e.g. we do not need a model to know people are forced to flee due to conflict). However, the model attempts to find the best combination that could predict the target variable, in this case: population flow numbers and geographical locations, in an specific period of time.

• Current number of Persons of Concern (PoC) per region per month

• Departures number of PoC per region per month

• Arrivals number of PoC per region per month

• Number of fatalities per region per month

• Number of violent conflict incidents per region per month

• River levels per river per month

• River discharge per river per month

• Rainfall average per region per month

• Water drum price per region per month

• Local goat Price per region per month

• The data behind the engine is anonymized, aggregated per month and per region and complies with UNHCR Data Protection policy standards

• The models used for each region in the engine represent the top 3 best ‘fit’ functions that can explain the behaviour of seven years of historical data. The target variable (y) is number of arrivals for each region.

• The model is trained with data until 30 June 2017 and the date that the predictive algorithm was built is 1 September 2017.