The UN Mine Action Service (UNMAS) is an entity within the United Nations’ Department of Peacekeeping Operations that works to eliminate the threat posed by mines and other explosives. Clearing explosives paves the way for economic development, but measuring this impact has proved a challenge for the organization.
In fall 2021, UNMAS tasked MIT Sloan’s Analytics Lab (A-Lab) with developing a machine learning model that could use satellite imagery to detect the presence of buildings and roads—key indicators of development—in areas of Afghanistan that UNMAS had cleared of explosives.
Four students from MIT Sloan’s Master of Business Analytics program teamed up to address this challenge: Alexander Birch, Manik Mukherjee, Ultan O'Rourke, and Mariana Suarez. The team applied their experience in data science as well as a passion for international development. “Seeing the impact that our contribution could have toward the United Nations’ work on a global scale helped us stay motivated throughout the different challenges we faced,” O’Rourke says.
The team began by meeting with a UNMAS team member to learn what buildings and roads look like in Afghanistan. The students then manually labeled hundreds of images to develop training data for their machine learning model. Ultimately, the team produced a proof-of-concept model capable of attaining an accuracy rate greater than 96 percent in a test scenario.
The students say the Action Learning project gave them an invaluable chance to run and execute a data science project from start to finish. “One of our key takeaways was the importance of effective communication,” O’Rourke says. “Being able to understand the entirety of the problem from the beginning, communicate your plans and progress to stakeholders in a nontechnical way, and present your results as actionable insights are key criteria to creating impact with data.”
Rory Collins, global information management and analytics advisor at the UN Office for Project Services, says working with A-Lab was a positive experience for him and for his fellow hosts on the UNMAS project. “The students were extremely professional, very agile to our evolving requirements, and we felt very engaged by their contribution and enthusiasm,” he says.
Plus, the machine learning models the students developed were immediately applicable to the satellite imagery analysis under way at UNMAS, he says. “The issue we are facing now isn't if we implement the work from the students, it's how we decide which of the many applications of their work we prioritize.”