Susceptibility assessment and zoning of coastal landslides based on heterogeneous ensemble machine learning models

ObjectiveWith the rapid development of marine engineering and the increasing frequency of extreme weather events, the risk of coastal landslides has risen significantly. However, existing studies on landslide susceptibility and zoning mainly focus on inland mountainous landslides, and systematic research on coastal landslide susceptibility remains insufficient. MethodsIn this study, the coastal zone of Fujian Province was selected as the study area. Historical data on coastal landslides were collected, and a susceptibility assessment index system suitable for coastal landslides was established using the information gain ratio method and Pearson correlation coefficient method. Particle swarm optimization support vector machine (PSO-SVM) and random forest (RF) were used as base learners to construct a stacking heterogeneous ensemble learning model. This model was adopted to perform the susceptibility assessment and zoning of coastal landslides in Fujian Province, and the influence of different training-to-testing data splitting ratios on the prediction accuracy of the heterogeneous ensemble model was also analyzed. ResultsThe comparison results demonstrated that the Stacking model performed optimally when the training-to-testing ratio was 70:30, achieving Accuracy of 0.869, Precision of 0.842, Recall of 0.909, and F1-Score of 0.874. Compared with other models, the Accuracy, Precision, and F1-Score improved by up to 0.198, 0.227, and 0.140, respectively. In addition, the area under the curve (AUC) value was 0.938, 0.019−0.216 higher than that of the other models. ConclusionThe findings indicate that the Stacking heterogeneous ensemble model exhibits strong applicability and excellent performance in susceptibility assessment of coastal landslides.