Susceptibility Model Research Articles

Developing a hybrid deep learning model with explainable artificial intelligence (XAI) for enhanced landslide susceptibility modeling and management

Developing a hybrid deep learning model with explainable artificial intelligence (XAI) for enhanced landslide susceptibility modeling and management

A novel CGBoost deep learning algorithm for coseismic landslide susceptibility prediction

The accurate prediction of landslide susceptibility shortly after a violent earthquake is quite vital to the emergency rescue in the 72-h “golden window”. However, the limited quantity of interpreted landslides shortly after a massive earthquake makes landslide susceptibility prediction become a challenge. To address this gap, this work suggests an integrated method of Crossing Graph attention network and xgBoost (CGBoost). This method contains three branches, which extract the interrelations among pixels within a slope unit, the interrelations among various slope units, and the relevance between influencing factors and landslide probability, respectively, and obtain rich and discriminative features by an adaptive fusion mechanism. Thus, the difficulty of susceptibility modeling under a small number of coseismic landslides can be reduced. As a basic module of CGBoost, the proposed Crossing graph attention network (Crossgat) could characterize the spatial heterogeneity within and among slope units to reduce the false alarm in the susceptibility results. Moreover, the rainfall dynamic factors are utilized as prediction indices to improve the susceptibility performance, and the prediction index set is established by terrain, geology, human activity, environment, meteorology, and earthquake factors. CGBoost is applied to predict landslide susceptibility in the Gorkha meizoseismal area. 3.43% of coseismic landslides are randomly selected, of which 70% are used for training, and the others for testing. In the testing set, the values of Overall Accuracy, Precision, Recall, F1-score, and Kappa coefficient of CGBoost attain 0.9800, 0.9577, 0.9999, 0.9784, and 0.9598, respectively. Validated by all the coseismic landslides, CGBoost outperforms the current major landslide susceptibility assessment methods. The suggested CGBoost can be also applied to landslide susceptibility prediction in new earthquakes in the future.

Modelling on assessment of flood risk susceptibility at the Jia Bharali River basin in Eastern Himalayas by integrating multicollinearity tests and geospatial techniques

AbstractClimate change and anthropogenic factors have exacerbated flood risks in many regions across the globe, including the Himalayan foothill region in India. The Jia Bharali River basin, situated in this vulnerable area, frequently experiences high-magnitude floods, causing significant damage to the environment and local communities. Developing accurate and reliable flood susceptibility models is crucial for effective flood prevention, management, and adaptation strategies. In this study, we aimed to generate a comprehensive flood susceptibility zone model for the Jia Bharali catchment by integrating statistical methods with expert knowledge-based mathematical models. We applied four distinct models, including the Frequency Ratio model, Fuzzy Logic (FL) model, Multi-criteria Decision Making based Analytical Hierarchy Process model, and Fuzzy Analytical Hierarchy Process model, to evaluate the flood susceptibility of the basin. The results revealed that approximately one-third of the Jia Bharali basin area fell within moderate to very high flood-prone zones. In contrast, over 50% of the area was classified as low to very low flood-prone zones. The applied models demonstrated strong performance, with ROC-AUC scores exceeding 70% and MAE, MSE, and RMSE scores below 30%. FL and AHP were recommended for application among the models in areas with similar physiographic characteristics due to their exceptional performance and the training datasets. This study offers crucial insights for policymakers, regional administrative authorities, environmentalists, and engineers working in the Himalayan foothill region. By providing a robust flood susceptibility model, the research enhances flood prevention efforts and management, thereby serving as a vital climate change adaptation strategy for the Jia Bharali River basin and similar regions. The findings also have significant implications for disaster risk reduction and sustainable development in vulnerable areas, contributing to the global efforts towards achieving the United Nations' Sustainable Development Goals.

Landslide Susceptibility Mapping of Chaliyar River Basin by Multivariate Statistical Model

Landslides were frequently observed in nature that can result in significant property damage and fatalities. Land management in landslide-prone areas can be aided by preparing a landslide susceptibility map. The landslide susceptibility of Chaliyar river basin was evaluated using the logistic regression (LR) technique. For this, an inventory map of 592 prior landslides was created using Landsat 8 OLI satellite imagery. The inventory of landslides was then randomly split into 30% and 70% for model training and validation respectively. Fifteen landslide causative factors viz., Slope, Aspect, Curvature, Relative Relief, TWI, Distance to Road, Distance to Streams, Distance to Lineaments, Land Use Land Cover, Drainage Density, Road Density, Lineament Density, Geomorphology, Soil Texture, NDVI were considered for landslide susceptibility modelling. Utilising a Receiver Operating Characteristics Curve (ROC) and Area Under Curve (AUC) value, the resulting susceptibility maps were validated. Analysis reveals that the validation stage of the LR model had a ROC-AUC value of 0.815. The study also demonstrates that slope, soil texture and LULC play a substantial role on the occurrence of landslides in the study area. The proposed landslide susceptibility model is appropriate, taking into account the ROC-AUC (0.815), and can be applied to future land use planning and landslide mitigation in the Chaliyar basin.

Integrating deep learning neural network and M5P with conventional statistical models for landslide susceptibility modelling

Integrating deep learning neural network and M5P with conventional statistical models for landslide susceptibility modelling

Landslide susceptibility mapping based on the reliability of landslide and non-landslide sample

Spatial data sampling can improve the performance in geo-spatial prediction. However, measuring the reliability of polygon-based data in sampling process is still a challenge. In this study, a reliability-based sampling (RBS) method was proposed to deal with this question and it was applied in landslide susceptibility mapping. First, the prototype of landslide was extracted from landslide polygon data, then, the reliability of landslide samples and non-landslide samples is measured using the similarity in environmental factor between the candidate samples and the prototype. The mutual exclusion reliability threshold setting method is used to collect the landslide samples and non-landslide samples with reliability over certain threshold. A case study demonstrates that the RBS method is better than existing representative method (i.e. Landslide entity) in terms of Accuracy and AUC with different sample sizes. In summary, The RBS is an efficient method to improve the spatial pattern of samples can also be applied to in other geo-spatial predictions.

Modeling and Evaluation of the Susceptibility to Landslide Events Using Machine Learning Algorithms in the Province of Chañaral, Atacama Region, Chile

Landslides represent one of the main geological hazards, especially in Chile. The main purpose of this study is to evaluate the application of machine learning algorithms (SVM, RF, XGBoost and logistic regression) and compare the results for the modeling of landslide susceptibility in the province of Chañaral, III region, Chile. A total of 86 sites are identified using various sources, in addition to 86 non-landslide sites. This spatial data management and analysis are conducted using QGIS software. The sites are randomly divided, and then a cross-validation process is applied to calculate the accuracy of the models. After that, from 22 conditioning factors, 12 are chosen based on the information gain ratio (IGR). Subsequently, five factors are excluded by the correlation criterion. After this analysis, two indices not previously utilized in the literature, the NDGI (normalized difference glacier index) and EVI (enhanced vegetation index), are employed for the final model. The performance of the models is evaluated through the area under the ROC (receiver operating characteristic) curve (AUC). To study the statistical behavior of the model, the Friedman nonparametric test is performed to compare the performance with the other algorithms and the Nemenyi test for pairwise comparison. Of the algorithms used, RF (AUC = 0.957) and XGBoost (AUC = 0.955) have the highest accuracy values measured in AUC compared to the other models and can be used for the same purpose in other geographic areas with similar characteristics. The findings of this investigation have the potential to assist in land use planning, landslide risk reduction, and informed decision making in the surrounding zones.

Forecasting daily COVID-19 cases with gradient boosted regression trees and other methods: evidence from U.S. cities.

There is a vast literature on the performance of different short-term forecasting models for country specific COVID-19 cases, but much less research with respect to city level cases. This paper employs daily case counts for 25 Metropolitan Statistical Areas (MSAs) in the U.S. to evaluate the efficacy of a variety of statistical forecasting models with respect to 7 and 28-day ahead predictions. This study employed Gradient Boosted Regression Trees (GBRT), Linear Mixed Effects (LME), Susceptible, Infectious, or Recovered (SIR), and Seasonal Autoregressive Integrated Moving Average (SARIMA) models to generate daily forecasts of COVID-19 cases from November 2020 to March 2021. Consistent with other research that have employed Machine Learning (ML) based methods, we find that Median Absolute Percentage Error (MAPE) values for both 7-day ahead and 28-day ahead predictions from GBRTs are lower than corresponding values from SIR, Linear Mixed Effects (LME), and Seasonal Autoregressive Integrated Moving Average (SARIMA) specifications for the majority of MSAs during November-December 2020 and January 2021. GBRT and SARIMA models do not offer high-quality predictions for February 2021. However, SARIMA generated MAPE values for 28-day ahead predictions are slightly lower than corresponding GBRT estimates for March 2021. The results of this research demonstrate that basic ML models can lead to relatively accurate forecasts at the local level, which is important for resource allocation decisions and epidemiological surveillance by policymakers.

In vivo dose response and efficacy of the β-lactamase inhibitor, durlobactam, in combination with sulbactam against the Acinetobacter baumannii-calcoaceticus complex.

Multi-drug resistant (MDR) Acinetobacter baumannii is emerging as a pathogen of increasing prevalence and concern. Infections associated with this Gram-negative pathogen are often associated with increased morbidity and mortality and few therapeutic options. The β-lactamase inhibitor sulbactam used commonly in combination with ampicillin demonstrates intrinsic antibacterial activity against A. baumannii acting as an inhibitor of PBP1 and PBP3, which participate in cell wall biosynthesis. The production of β-lactamases, particularly class D oxacillinases, however, has limited the utility of sulbactam resorting to increased doses and the need for alternate therapies. Durlobactam is a non-β-lactam β-lactamase inhibitor that demonstrates broad β-lactamase inhibition including class D enzymes produced by A. baumannii and has shown potent in vitro activity against MDR A. baumannii, particularly carbapenem-resistant isolates in susceptibility and pharmacodynamic model systems. The objective of this study is to evaluate the exposure-response relationship of sulbactam and durlobactam in combination using in vivo neutropenic thigh and lung models to establish PK/PD exposure magnitudes to project clinically effective doses. Utilizing established PK/PD determinants of %T>MIC and AUC/MIC for sulbactam and durlobactam, respectively, non-linear regressional analysis of drug exposure was evaluated relative to the 24-hour change in bacterial burden (log10 CFU/g). Co-modeling of the data across multiple strains exhibiting a broad range of MIC susceptibility suggested net 1-log10 CFU/g0 reduction can be achieved when sulbactam T>MIC exceeds 50% of the dosing interval and durlobactam AUC/MIC is 10. These data were ultimately used to support sulbactam-durlobactam dose selection for Phase 3 clinical trials.

A hybrid one-vertex model for susceptible–infected–susceptible diseases on networks with partial connection information

We analyze the propagation of the susceptible–infected–susceptible model on the network, where the partial connection information is known. In order to analyze the spread of network epidemics, the quenched mean-field method requires complete connection information, while the heterogeneous mean-field method requires complete topology information. When only partial connection information is available, the quenched mean-field method may be disabled. In view of this situation, a hybrid model consisting of connection and topology information is proposed. The theoretical results are in good agreement with stochastic simulations on scale-free networks. In addition, as the proportion of connection information in the network increases, model predictions about final epidemic size become more accurate when the effective transmission rate is slightly larger than the simulation threshold.

Tuning the magnetic properties of doped ZnS using transition metal doping: A multi-scale computational approach

Transition metal (TM)-doped dilute magnetic semiconductors (DMS) have emerged as promising materials for spintronic applications such as spin-polarized transport and information storage. However, achieving robust room-temperature ferromagnetism in DMS remains a key challenge. This work undertakes a comprehensive investigation of the effects of Ti- and Cr-doping and (Ti,Cr)-codoping on the structural, electronic, and magnetic properties of wurtzite ZnS using first-principles pseudopotential plane-wave self-consistent field calculations and Monte Carlo simulations. The first-principles DFT calculations were performed using the spin-polarized GGA+U approach to account for strong electron–electron interactions. Structural relaxations revealed expanded lattice parameters and bond lengths for the doped systems compared to pure ZnS, attributed to the larger ionic radii of the dopants. Formation energy calculations confirmed the thermodynamic stability of doping. The band structure and density of states analysis demonstrated the half-metallic nature of the doped ZnS systems with 100% spin polarization induced by strong hybridization between the TM-3d and S-3p states. This mediates robust ferromagnetic coupling, with total magnetic moments around 4-8 μB/cell depending on dopant type and concentration. To complement the zero-temperature DFT results, temperature-dependent Monte Carlo simulations using the single-spin flip Heat Bath algorithm were implemented. The modeling of magnetization, susceptibility, specific heat, and hysteresis loops enabled extracting Curie temperatures up to 427.6 K for 12.5% Cr-doped ZnS. The synergistic combination of first-principles DFT and Monte Carlo computational techniques provides significant insights into tailoring the magnetic properties of TM-doped ZnS dilute magnetic semiconductors. The tunability of structural, electronic, and magnetic characteristics through control of dopant selection and concentration demonstrates the promising potential of transition metal-doped ZnS for spintronic devices operable at room-temperature.

The role of oxidative stress levels and S100B levels in children with functional neurological disorder

Background/Aim: Although stressors and traumatic life events are known to be predisposing factors for developing functional neurological disorder (FND), the etiology of the disorder has not been fully elucidated. In this study, oxidative stress parameters and serum levels of S100B protein were investigated in pediatric patients with functional neurological disorder. The association of these parameters with stress factors and traumatic life events was investigated. Methods: This case-control study included a control group of 35 patients aged 8-18 years and 32 healthy subjects diagnosed with functional neurological disorder. The Childhood Trauma Scale and Dissociative Experiences Scale were applied to both groups. Serum levels of the patient and control groups were compared by blood sampling: total oxidant status (TOS), total antioxidant status (TAS), oxidative stress index (OSI), and S100B. Results: It was found that the Childhood Trauma Scale total score, emotional neglect, emotional abuse, physical neglect and the Dissociative Experiences Scale were statistically significantly higher in the FND group than in the control group (P˂0.001, P˂0.001, P=0.013, P=0.017, P˂0.001). Stressors were found to be statistically significantly higher in the FND group than in the control group (P=0.020). There was no statistical difference between the FND group and control groups regarding the TAS, TOS, OSI, and S100B levels (P=0.965, P=0.228, P=0.268, P=0.517, respectively). Conclusion: Our study is the first to investigate TOS, TAS, OSI and S100B in children with FND. In our study, although stressors and traumatic experiences were significantly higher in the functional neurological disorder group compared to the control group according to the stress susceptibility model, contrary to expectations, there was no significant difference in oxidative stress parameters and serum S100B levels. It was thought that the interaction between FND, which is characterized by biopsychosocial interaction and can manifest itself with various clinical symptoms, and stress may not be linear as initially thought, and the interaction between genetic predisposition and environmental factors may play a more complex role. The absence of significant differences observed in oxidative stress parameters and serum S100B levels may suggest that we should focus on different pathways and different potential biomarkers that need to be investigated in the future to understand the etiology and diagnosis of FND. However, the limitations noted above may affect the generalizability of the study findings.

Assessing the imperative of conditioning factor grading in machine learning-based landslide susceptibility modeling: A critical inquiry

Current machine learning approaches to landslide susceptibility modeling often involve grading conditioning factors, a method characterized by substantial subjectivity and randomness. The necessity and rationality of such grading have sparked continued debate. Recognizing the potential profound impact of this grading on the results of models, we conducted an in-depth study focusing on four townships within the Wanzhou section of the Three Gorges Reservoir area. A comprehensive assessment was conducted using three traditional machine learning models, five ensemble learning models, and four deep learning models to evaluate the implications of continuous factor grading. Three grading strategies were explored: non-grading, equal intervals, and natural breaks. Further investigation was conducted to determine how various grade levels (e.g., 4, 6, 8, 12, 16, 20) affect model efficacy. Our analysis reveals that the Support Vector Machine (SVM) model performs optimally with an 8-level grading using natural breaks. In contrast, a decision tree (DT) and its associated ensemble models are more effective without grading. For Multi-Layer Perceptron Neural Network (MLPNN) and Convolutional Neural Networks (CNN) models, a natural breaks grading exceeding 8 levels is advisable. Gated Recurrent Unit (GRU) and Deep Neural Networks (DNN) models benefit from an equidistant grading strategy of over 12 levels, while Long Short-Term Memory Neural Networks (LSTM) models thrive with an equidistant grading surpassing 16 levels. This study is pioneering in introducing grading guidelines for machine learning models in landslide susceptibility modeling. Our findings offer invaluable insights for future research, setting a path towards more standardized practices in this field. This enhances the bridge between theoretical knowledge and its real-world application, promoting a more rigorous and systematic grading approach and advancing the standardization of landslide susceptibility modeling.

A REVIEW OF STATISTICAL APPROACHES USED FOR LANDSLIDE SUSCEPTIBILITY ANALYSIS WITH THE HELP OF REMOTE SENSING AND GIS TECHNOLOGY

One of the most significant environmental risks is landslides. It might happen due to human action or natural occurrence and have significant social, economic, and environmental repercussions. The danger of landslides occurring in a certain area based on regional topographical features is known as landslide susceptibility. Numerous articles evaluating landslide susceptibility have been written since the middle of the seventies utilizing a variety of techniques and methodologies in diverse geographical and environmental situations.To critically review this literature, major factors that are considered for landslide susceptibility, and the models that are used for landslide susceptibility analysis. Then these data are represented in a graphical visualization form.Landslide susceptibility zoning, The goal of the current study, which was undertaken as a review, was to provide a critical analysis of statistical methods used for landslide susceptibility modelling and associated terrain zonation from 2010 to 2023.There has been an increase in the number of publications from 2019 onwards on the topic of landslide susceptibility. In 20.8% of the studies, the FR model and LR model has been chosen as the most popular approach for determining landslide vulnerability. Other then Logistic regression and Frequency ratio model, other models like Fuzzy, WoE, ANN, SVM and RF are also popular for landslide susceptibility. As the technology in the field of GIS&RS is updating day to day and recent policy developments, are significant factors in boosting interest in landslide susceptibility research.

Spatial Analysis of Causative Factors for Landslide Susceptibility on Java Island

Landslide is the most frequent natural disaster on Java Island with around 7300 occurrences in 1998-2023 (BNPB), causing 1807 fatalities and 45770 infrastructure damages. Landslide susceptibility modeling can be implemented as a basis for landslide risk modeling in Java Island for mitigation purposes. This research aims to model landslide susceptibility in Java Island and validate it based on historical landslide occurrence data. The methodology used for the landslide susceptibility model is based on the relationship between the distribution of landslide events and the causative factors in each of their classification class. Meanwhile, the methodology used for validation is Receiver Operating Characteristic (ROC) curve analysis, conducted by comparing the landslide susceptibility model with landslide occurrence data to obtain the area under the curve (AUC) value that shows the performance of the model. In this study, historical landslide occurrence data from two different sources, NASA and BNPB, were used to generate two separate models to see the difference of performance between both models. Each landslide occurrence data from both sources is divided into two parts; 70% of it is used to develop landslide susceptibility map and the rest is used for validation process. As many as ten causative factors were used to generate the model; elevation, slope, aspect, lithology, land cover, rainfall, river density, PGA (Peak Ground Acceleration), and distance to fault and river. The results show that based on landslide susceptibility map from BNPB data, Java Island is dominated with low susceptibility, that is about 58447 km2. Meanwhile, based on landslide susceptibility map from NASA data, Java Island is dominated by medium susceptibility, that is around 76731 km2. The performance for the models based on AUC values are 0.98-0.99 successful for NASA’s dataset and 0.91-0.92 successful for BNPB’s dataset in assessing landslide susceptibility in Java Island.

Flood susceptibility mapping at the country scale using machine learning approaches

AbstractRiver (fluvial), surface water (pluvial) and coastal flooding pose a significant risk to the United Kingdom. Therefore, it is important to assess flood risk particularly as the impacts of flooding are projected to increase due to climate change. Here we present a high resolution combined fluvial and pluvial flood susceptibility map of England. This flood susceptibility model is created by using past flood events and a series of meaningful hydrological parameters to a training machine learning model. We tested the relative performance of different machine learning algorithms, including Classification and Regression Trees, Random Forest and XGBoost and found the XGBoost performed the best, with an area under the receiver operating characteristic ROC Curve (AUC) of 0.93. We also found the model performed well on unseen areas, and we discuss the possibility of extending to regions that has no information on past flood events. Additionally, to aid in understanding what factors may impact flood risk to a particular area, we used Shapley additive explanations which allowed us to investigate the sensitivity of the predicted flood probability to flood factors at a given location.

Longitudinal Relationships Among Maternal Depression, Social Parenting, and Young Children’s Externalizing Behavior Problems: Moderated Mediation Effect of Young Children’s Negative Emotionality

Objectives: This study investigated whether young children’s negative emotionality moderates the longitudinal mediating effect of maternal social parenting on the relationship between maternal depression and young children’s externalizing behavioral problems.Methods: The 2008 (T1)~2012 (T5) data of the Panel Study on Korean Children (PSKC) were used. Participants were 1703 children aged 0 to 4 years old, their mothers, and their fathers. Descriptive statistics were performed for the variables and Pearson’s correlation analyses were carried out to identify the association with the variables. SPSS PROCESS macro version 4.0 software was used for analyzing the mediating effect (Model 4) and moderated mediation effect (Model 14). The significance of the indirect effect was verified via a bootstrapping test.Results: According to the results, first, maternal social parenting (T4) had a partial mediating effect on the relationship between maternal depression (T1-T3) and young children’s externalizing behavioral problems (T5). Second, young children’s negative emotionality (T4) had a moderating effect on the relationship between maternal social parenting(T4) and young children’s externalizing behavioral problems (T5). Third, the mediating effect of maternal social parenting (T4) on the relationship between maternal depression (T1-T3) and young children’s externalizing behavioral problems (T4) was moderated by young children’s negative emotionality (T4). Specifically, the indirect effect of maternal depression on young children’s externalizing behavior problems through maternal social parenting was more impactful among young children with average or high on negative emotionality. The findings supported a diathesis-stress model rather than a differential susceptibility model regarding the interaction effect between young children’s temperament and maternal parenting on young children’s externalizing behavior problems.Conclusion:In the context of the diathesis-stress model, adjusting interventions to young children’s negative emotionality, which serves as a vulnerability for negative parenting environments, may be a useful way to prevent their externalizing behavior problems.

Does early child negative emotionality moderate the association between maternal stimulation and academic readiness and achievement?

We investigated whether child temperament (negative emotionality, 5 months) moderated the association between maternal stimulation (5 months-2½ years) and academic readiness and achievement (vocabulary, mathematics, and reading). We applied structural equation modeling to the data from the Quebec Longitudinal Study of Child Development (N = 1121-1448; mostly Whites; 47% girls). Compared to children with low negative emotionality, those with high negative emotionality had higher levels of academic readiness (6 years) and mathematics achievement (7 years) when exposed to high levels of maternal stimulation (β = 3.17, p < .01 and β = 2.91, p < .01, respectively). The results support the differential susceptibility model whereby highly emotionally negative children were more susceptible to the influences of low and high levels of maternal stimulation in academic readiness and mathematics achievement's developments.

A New Lithospheric Density and Magnetic Susceptibility Model of Iran, Starting From High‐Resolution Seismic Tomography

AbstractWe propose a new model for the crust and upper mantle in Iran by joint inversion of gravity and magnetic fields, constrained with a seismic tomography model. We then calculate shear modulus from the Vs velocities and densities. The crust and mantle tomography model is first converted to a density cube through empirical and petrological velocity‐density relations. The starting susceptibility is assigned to a two‐layer homogeneous model, above a heat flow‐derived Curie depth. Considering the uncertainties in the density‐velocity relations, and the starting layered susceptibility variation, we refine the model by a constrained inversion of the gravity and magnetic fields with a Bayesian approach, producing the final 3D density and susceptibility model. The area is tectonically active with high seismicity and active faulting which are regulated by the crustal density and rigidity variations. Higher rigidity matches lower seismicity and extended deserts and basins, suggesting the control of their development. The Neo‐Tethys suture, extending ∼1,500 km long, as well as the Paleo‐Tethys suture match crustal scale density variations, defining characteristics of the lighter Arabian plate and denser Eurasian crust. The South Caspian Basin is enigmatic, due to focusing on the seismicity along all its borders, but with relatively low average rigidity, which is contrary to what is observed for Iran, where the reduced rigidity correlates with higher seismicity. The 3D density model will be useful for numerical geodynamic models and obtaining geologic inferences from the crustal‐scale units.

A Comparative Study of Landslide Susceptibility Mapping Using Bagging PU Learning in Class-Prior Probability Shift Datasets

Landslide susceptibility mapping is typically based on binary prediction probabilities. However, non-landslide samples in modeling datasets are often unlabeled data, and the phenomenon of class-priori shift, that is, the proportion of landslide samples frequently deviates from real-world scenarios and is spatially heterogeneous. By comparing the classification performance and predicted probability distributions across multiple unbalanced datasets with known and unknown sample proportions, this study assesses the landslide susceptibility model’s generalization ability in the context of class-prior shifts. The study investigates the potential of Bagging PU Learning, a semi-supervised learning approach, in improving the generalization performance of landslide susceptibility models and proposes the Bagging PU-GDBT algorithm. Our findings highlight the effectiveness of Bagging PU Learning in enhancing the recall of landslides and the generalization capabilities of models on unbalanced datasets. This method reduces prediction uncertainties, especially in high and very high susceptibility zones. Furthermore, results emphasize the superiority of models trained on balanced datasets with 1:1 sample ratio for landslide susceptibility mapping over those trained on unbalanced datasets.