Evaluating livelihoods interventions and predicting needs through data science and machine learning
Challenge
Despite UNHCR’s recent efforts to mainstream data-driven analysis and decision-making, many data-driven projects are still siloed within specialist units, rather than being operationalised to inform field-based work. UNHCR technical units in the field also continue to collect large amounts of data that may never be analysed beyond a very basic level, if at all.
Solution
Using data science and machine learning techniques to pilot deeper analysis of data collected by humanitarian organization HIAS on the progress of displaced households included in a specific livelihoods intervention. Using machine learning, the team will sort cases according to performance, identifying trends and common characteristics.
Impact
Data-driven prediction of which cases are most likely to need additional support from the start of livelihoods interventions will enable a more effective response and amplify successes within livelihood interventions.
Project impact
Other information
UNHCR Peru's Graduation Model lifts refugee and migrant families out of extreme poverty through sequenced support, but many do not complete the programme, leaving the most vulnerable still at risk. Funded by the Data Innovation Fund, this project applied machine learning (PCA, clustering, and logistic regression) to 210 families across two cohorts to identify what really drives successful graduation. The analysis revealed that family characteristics, such as GBV history, chronic illness, and lack of documentation, matter more than programme performance, while attending the financial workshop, having community support networks, and holding a Carné de Extranjería significantly increase graduation odds. The evidence is now being used to redesign targeting and differentiated support, and is replicable across Graduation Model operations in the region. Find out more about the project here.