Journal of Frontier Research across Disciplines
Download PDF
ABSTRACT
Introduction: Human migration significantly impacts social and economic systems, yet traditional forecasting models struggle to capture its complex, high-frequency dynamics. Advances in machine learning offer improved predictive capabilities, but most models remain static and limited in temporal and spatial integration. This study proposes a hybrid CNN-LSTM framework to predict migration flows by combining spatial and temporal feature extraction and integrating multi-source data for real- time prediction.
Methods: Using 114,612 records from the Dabat Health and Demographic Surveillance System (2008–2021), data were preprocessed, normalize, and dimension were analyzed using PCA analysis. Key predictors were identified via feature importance score analysis. A hybrid 1D CNN-LSTM model was trained using the Adam optimizer and performance was assessed with accuracy, precision, recall, F1 score, ROC-AUC, and confusion matrices.
Results: The model achieved a training accuracy of 0.9960, while both validation and test accuracies were approximately 0.9898, indicating consistently high performance. Precision, recall, and F1-scores followed a similar trend, with validation and test scores of 0.9897 and 0.9898, respectively, and minimal misclassification. Feature analysis identified End_type, Reason_for_migration, and Start_type as the dominant predictors. The dashboard effectively visualized migration trends, including in- and out-migrant counts, demographic distributions, and real-time hotspots. The result demonstrates that the robustness and reliability of the hybrid CNN-LSTM approach for migration prediction.
Conclusion: The hybrid CNN-LSTM approach delivers highly accurate, real-time migration forecasts and identifies key determinants. By combining spatial-temporal modeling with interactive visualization, the system supports data-driven decision-making and scalable migration monitoring in resource-limited contexts.
Keywords: Human Migration; Migration Prediction; Spatiotemporal Modeling; Machine Learning; Real- Time Monitoring; Multi-source Data