Enhanced Landslide Spatial Prediction Using Hybrid Deep Learning Model and SHAP Analysis: A Case Study of the Tuyen Quang-Ha Giang Expressway, Vietnam

This study presents an approach to improve landslide spatial prediction along the Tuyen Quang-Ha Giang (TQ-HG) Expressway in northern Vietnam by employing advanced deep learning techniques, specifically Artificial Neural Networks (ANN), Convolutional Neural Networks (CNN), Deep Neural Networks (DNN), and a novel hybrid model, the CNN-DNN model. Utilizing a dataset consisting of 139 landslide occurrences along the Expressway and fourteen influential variables, the study demonstrates the superior predictive performance of the hybrid CNN-DNN model across multiple evaluation metrics, including sensitivity (86.96%), specificity (90%), accuracy (88.37%), Kappa statistic (0.767), Root Mean Square Error (RMSE) (0.341), and Area Under the Curve (AUC) (0.93). Additionally, employing the Shapley Additive Explanations (SHAP) method identifies key factors affecting landslide susceptibility, with maximum rainfall emerging as the most influential factor, followed by average annual rainfall, Normalized Difference Vegetation Index (NDVI), distance to roads, and elevation. This study represents a significant advancement in accurate road-side landslide susceptibility prediction, with potential applications in similar geographic regions facing landslide hazards.