Journal
LAND
Volume 10, Issue 12, Pages -Publisher
MDPI
DOI: 10.3390/land10121370
Keywords
predictive modelling; correlation analysis; logistic regression; random forest; landslide susceptibility; western carpathians
Categories
Funding
- Projects: VEGA Historical and present changes in the landscape diversity and biodiversity caused by natural and anthropogenic [2/0132/18]
Ask authors/readers for more resources
This study evaluated landslide susceptibility in the northwestern part of the Kysuca River Basin in Slovakia using logistic regression and random forest models. The random forest model outperformed the logistic regression model in all measures, achieving very good results on validation datasets. The importance of assessing the calibration of predicted probabilities in landslide susceptibility modeling was also highlighted in the results of the logistic regression model.
Landslides are the most common geodynamic phenomenon in Slovakia, and the most affected area is the northwestern part of the Kysuca River Basin, in the Western Carpathian flysch zone. In this paper, we evaluate the susceptibility of this region to landslides using logistic regression and random forest models. We selected 15 landslide conditioning factors as potential predictors of a dependent variable (landslide susceptibility). Classes of factors with too detailed divisions were reclassified into more general classes based on similarities of their characteristics. Association between the conditioning factors was measured by Cramer's V and Spearman's rank correlation coefficients. Models were trained on two types of datasets-balanced and stratified, and both their classification performance and probability calibration were evaluated using, among others, area under ROC curve (AUC), accuracy (Acc), and Brier score (BS) using 5-fold cross-validation. The random forest model outperformed the logistic regression model in all considered measures and achieved very good results on validation datasets with average values of AUC(val)=0.967, Acc(val)=0.928, and BSval=0.079. The logistic regression model results also indicate the importance of assessing the calibration of predicted probabilities in landslide susceptibility modelling.
Authors
I am an author on this paper
Click your name to claim this paper and add it to your profile.
Reviews
Recommended
No Data Available