Can Level-2 Firth’s Bias-reduced logistic regression be considered a robust approach for predicting landslide susceptibility?

The implementation of effective landslide mitigation strategies relies heavily on the availability of accurate and reliable landslide susceptibility map. This study focuses on the adequacy evaluation of the Level-2 Firth’s Bias-Reduced Logistic Regression (BLR) to predict landslide susceptibility. The study was performed at the mountain Seunghak which lies in the southern-west part of Busan. A total of 57 multi-temporal landslides since 2006 to 2019 were identified and plotted in geographic information system (GIS) environment. Although, twelve spatial environmental variables were selected for the analysis, topographic wetness index was removed to avoid a collinearity issue. The dataset was randomly divided into two sets: training set (70%) and test set (30%), ensuring they did not overlap. In order to assess the performance of the model, two different cross-validation methods i.e. random cross-validation (RCV) and spatial cross-validation (SCV) were applied. The overall accuracy was examined using area under the curve of receiver operating characteristic curve (mean AUC of RCV = 0.965, mean AUC of SCV = 0.939). The true positive and true negative values depicted correctly which showed the excellent adequacy of the prediction of the landslide occurrences. Among eleven environmental variables, slope played a significant role in the result of landslide prediction. The susceptibility estimation component of BLR model outperformed a standard logistic regression (LR) model, which we used as a benchmark. LR is the most widely used classifier in landslide research, making it a key point of comparison. In our discussion, we explored the strength and weaknesses of the new modeling framework and its potential applicability in various domains. We highlighted both the specific considerations related to hazards and geomorphology, as well as the broad implications of its application. [ABSTRACT FROM AUTHOR]