<![CDATA[Previous studies in coastal Java have used Random Forest, K-Nearest Neighbors, and Least Absolute Shrinkage and Selection Operator regression for land cover dynamics and statistical analyses, with recent 2024 findings showing that Extreme Gradient Boosting outperforms RF in classification accuracy. However, there is a gap in the model’s ability to simultaneously integrate all built-up land indices using an ensemble stacking algorithm. This study aims to develop a new classification framework based on dynamic stacking ensembles to improve tsunami vulnerability assessment accuracy in the coastal area of Jember Regency, East Java. This research method develops a hybrid RF-XGBoost model by integrating DEM topographic parameters and the built-up index from Landsat 8 imagery (2023–2025), beginning with K-Means clustering to identify homogeneous spatial patterns prior to classification. The novelty of this research lies in the development of a dynamic stacking ensemble method that integrates algorithms (RF and XGBoost) as base learners with multinomial logistic regression as a meta-learner. Unlike conventional single-model methods, this methodology uniquely synergizes built-up land indices combined with integrated topographic analysis to extract physical vulnerability parameters more precisely. This result demonstrated increased prediction robustness, with a high accuracy of 0.976 and an AUC-ROC of 0.5493, which significantly exceeds the class discrimination capabilities of the Random Forest or Extreme Gradient Boosting models alone. The contribution of this research is to propose an ensemble model that enhances predictive resilience and provides a reliable decision-support framework for tsunami risk mitigation planning.]]>
Copyrights © 2026
