| Abstrakt: |
Chin State in Myanmar experiences frequent landslides annually. This research aimed to construct GIS-based landslide susceptibility maps (LSMs) with two kinds of machine learning models, namely random forest (RF) and support vector machine (SVM). Firstly, a landslide inventory map was constructed by containing 213 landslide locations and randomly chosen 213 non-landslide locations; these location points were randomly divided into the training set (70 %) for the landslide susceptibility prediction model and the testing set (30 %) for the model validation. Secondly, twenty-one landslide conditioning factors were selected, and frequency ratio analysis was used to evaluate the relationship between each class of factors and landslide occurrences. Then, landslide susceptibility prediction modeling by RF and SVM models. Finally, the performance of the two models was evaluated with performance metrics (precision, recall, F1-Score, and accuracy), receiver operating characteristic (ROC) curves, and area under the ROC curve (AUC values). The RF model demonstrated superior performance across performance evaluation metrics, with a precision of 0.864, recall of 0.919, F1-Score of 0.891, and an accuracy of 0.894 on the training set, compared to the SVM model's precision of 0.854, recall of 0.807, F1-Score of 0.830, and accuracy of 0.825. The model validation by the testing set further confirmed that the RF model showed a precision of 0.839, recall of 0.897, F1-Score of 0.867, and an accuracy of 0.871, while the SVM model had a precision of 0.839, recall of 0.839, F1-Score of 0.839, and an accuracy of 0.839. Also, the results of AUC values showed that the RF model (training set AUC = 0.94, testing set AUC = 0.92), and SVM model (training set AUC = 0.89, testing set AUC = 0.88), respectively. Hence, these two landslide susceptibility prediction models demonstrated satisfactory results and good accuracy for LSMs in this research area, and the LSM from the RF model is better than the SVM model according to performance metrics and AUC values results. The resulting maps provide useful information on the likelihood of landslide occurrence, facilitating decision-making in land use planning and disaster management. [ABSTRACT FROM AUTHOR] |
