Cross-border commuting is the backbone of economic integration but remains poorly measured by traditional surveys or administrative records, especially at fine spatial scales. In this paper, we introduce a fully data-driven, machine-learning framework to detect and classify daily transnational commuters on the basis of geolocated social media data. We focus on three European border regions—the Greater Region of Luxembourg, the Basque Country, and the Øresund Region—and obtain eighteen network and spatiotemporal features (e.g., trip frequency, origin and destination dwell times, and cross-border trips). First, we train and evaluate four state-of-the-art classifiers (CatBoost, XGBoost, random forest, and k-nearest neighbours) on k-fold cross-validation, achieving up to 98% overall accuracy with 78% recall and 70% precision for detecting cross-border home-work movements at the individual movement level. Second, we apply zero-shot transfer learning by using Luxembourg-learned models to the two other regions and evaluate the results with official reports and population data. Our approach offers a scalable, replicable methodology for mapping cross-border labour mobility, providing valuable insights for infrastructure planning, social and economic equity, and climate adaptation policies at the urban and regional scale.