Integration of Geoscientific Data Mining and GBM Hyperparameter Optimization to Improve the Accuracy of Landslide Hazard Prediction

Landslides are geological disasters characterized by the movement of soil or rock masses down a slope due to geotechnical instability and environmental factors. They significantly impact infrastructure, life safety, ecosystems, and local economies. Therefore, accurate and reliable predictions are essential in supporting disaster risk mitigation. This research integrates the Geoscientific Data Mining approach with hyperparameter optimization on the Gradient Boosting Machine (GBM) algorithm to improve the accuracy and stability of landslide hazard prediction. Thirteen geospatial parameters are used as predictor variables, including aspect, elevation, slope, curvature, plan curvature, profile curvature, precipitation, lithology, NDVI, NDWI, flow accumulation, and earthquake. GBM is selected for its ability to model non-linear relationships and complex interactions between features, even when there is no explicit linear correlation between causal variables. Through a decision tree-based ensemble learning approach, GBM builds predictive models incrementally, which enables learning of hidden patterns in heterogeneous geoscience data. The optimization process is performed using Random, Manual and Grid Search Cross-Validation to obtain the best combination of parameters, focusing on the balance between the accuracy and stability of the predictive model. The research results improved the accuracy from 55% to 80%. Evaluation using precision, recall, and f1-score indicated more accurate and stable model performance. In addition, the confusion matrix provides good classification information, while the ROC Curve provides a high AUC value. The cross-validation score plotting shows the stability and generalization ability of the prediction model. This research proves that integrating geospatial data mining and optimized GBM can be an effective approach to supporting mitigation systems and development planning in landslide-vulnerable areas.