An Ensemble Model of Logistic Regression, Naïve Bayes, and Adaboost for Assessing the Landslide Spatial Probability - Study Case: Phuoc Son, Quang Nam, Vietnam and Umyeon, Seoul, Korea

Authors:
Open Access
Publisher website
Google Scholar
Abstract
This research employed a combination of commonly used machine learning (ML) models to improve the accuracy of predicting landslide spatial probability. The study areas were Phuoc Son, Quang Nam, Vietnam, and Mt. Umyeon, Seoul, Korea. Four ML models, namely logistic regression (LR), Bernoulli Naïve Bayes (BNB), Gaussian Naïve Bayes (GNB), and Adaboost (AD), were initially utilized to assess the spatial probability of landslides. Subsequently, an ensemble learning model was employed, using the results from the four ML models as input data, to produce the final landslide spatial probability. Evaluation metrics, including the areas under curve (AUC), were employed to evaluate the success of all ML models in predicting the spatial probability of landslides. The classified landslide susceptibility maps were generated based on the landslide spatial probability maps, employing different classifiers. The statistical significance of these maps was confirmed through the application of appropriate statistical tests, such as the Chi-square test. Comparative analysis between the individual ML models and the combination model revealed that the proposed combination model exhibited greater accuracy in predicting landslide spatial probability than the individual ML models.

This publication has 43 references indexed in Scilit: