A graph-augmented dual-task forecasting framework using GAT-XGBoost for global ICT goods trade and export market opportunity identification

Introduction Identifying promising export destinations requires considering both the likelihood of establishing a bilateral trade relationship and the potential trade scale once that relationship is established. Methods This study develops a graph-augmented dual-task forecasting framework for global information and communication technology (ICT) goods trade using GAT-XGBoost. The framework integrates BACI bilateral ICT goods trade data under the HS 2012 classification for 2012–2024 with country characteristics, gravity variables, historical trade information, and network features. A strict chronological design uses 2014–2021 for training, 2022 for validation, and 2023–2024 for out-of-period testing. Results On the 2024 test set, the framework achieves an AUC of 0.9631 and an Average Precision of 0.9394 for trade-link prediction, while conditional log-flow prediction achieves an RMSE of 1.7711 and an R 2 of 0.8344. Ablation results indicate that historical bilateral trade provides the dominant predictive signal, while observed network measures and graph-derived representations contribute complementary information. Discussion By combining predicted trade-link probabilities with conditional trade magnitudes, the framework produces exporter-specific destination rankings and opportunity classifications. The resulting measures support systematic screening of potential ICT export markets rather than causal inference or guaranteed future trade outcomes.