Decision tree-based prediction of economic activities in Colombia's Non-Interconnected Zones

Many remote areas in Colombia, known as Non-Interconnected Zones (NIZs), experience persistent energy poverty that limits their social and economic development. With the recent availability of detailed survey data from IPSE, the government's energy planning agency, it is now possible to study these communities with greater accuracy. However, there remains a lack of data-driven analyses that can directly support policymaking for NIZs. This study addresses that gap by developing a predictive model for livelihoods in NIZs. A decision tree classifier was trained on IPSE survey data from approximately 12,441 households, incorporating factors such as geographical region, education level, and access to public services. The model achieved an accuracy of about 75.9% on unseen data. The most influential predictors of a household’s main economic activity were the local region (i.e., county), formal community status, and the availability of sanitation and water services. By identifying these factors, the model provides practical insights for policymakers, guiding targeted interventions—such as infrastructure investment or resource allocation—to strengthen livelihoods in NIZs.