| dc.description.abstract | Analysis and prediction of climate-driven geohazards, such as rainfall-induced landslides and slope failures, are becoming more challenging given the changing climate where extreme events are inevitable. The purpose of this master thesis is to evaluate the capacities of four machine learning models, including Multi-Layer Perceptron Neural Network (MLP), Random Forest (RF), Gradient Boosted Regression Tree (GBRT), Extreme Gradient Boosting (XGBoost), in landslide susceptibility analysis. For the purpose of comparative analysis, a numerical method, Transient Rainfall Infiltration and Grid-Based Regional Slope-Stability Model (TRIGRS), is applied in this thesis, and model performance is assessed as well. The study is carried out based on a preliminary survey of rainfall-induced landslides near Kvam village, Norway, in June 2011. A methodology framework of landslide susceptibility modeling is proposed. Four critical data processing approaches, including feature selection, data resampling, data splitting, and feature scaling, are introduced and evaluated to improve the accuracy and efficiency of landslide-prone area prediction. In addition, the optimal hyperparameter optimization method is determined as Bayesian optimization by performing time efficiency analysis with the Grid Search method. As for model performance, the conclusion is drawn that GBRT is the optimal method for landslide risk assessment in the study case of Kvam according to seven popular model evaluation metrics. Other tree-based machine learning algorithms (RF and XGBoost) also show an overall outstanding performance and computational efficiency in comparison to the classical neural network model, MLP, and a numerical method, TRIGRS. The landslide susceptibility maps provided by five models are also analyzed statistically. The same result of model performance ranks is obtained, and over 95% of the total landslide release area is considered to have a moderate to very high probability for landslide movements according to the GBRT model in the study. | |