Abstract

Landslides pose a significant risk in the mountainous regions of Arunachal Pradesh, India.
Accurate prediction of landslide susceptibility is crucial for effective risk mitigation and land-use
planning. In recent years, machine learning (ML) models have shown promise in landslide
susceptibility mapping. This study compares the performance of multiple ML models, including
Radial Basis Neural Network (RBFN), Gradient Boosting, Logistic Regression (LR), and
Random Forest (RF), in predicting landslide susceptibility in Arunachal Pradesh. The report
focuses on feature selection using mutual information to enhance the performance of advanced
ML models and determine the accuracy of landslide susceptibility. The achieved accuracies for
different models are as follows: XGboost (0.58), SVM (0.64), Random Forest (0.58), RBFN
(0.70), and LR (0.76). Evaluation metrics such as accuracy and ROC-AUC curves are used to
assess the models' performance in distinguishing between landslide and non-landslide areas. The
study's findings contribute to the advancement of ML models for landslide susceptibility
mapping in Arunachal Pradesh. Policymakers, researchers, and practitioners can utilize this
information to develop targeted mitigation strategies, formulate land-use plans, and make
informed decisions to minimize the impact of landslides. Future studies may explore ensemble
techniques and incorporate additional data sources to further enhance the precision and reliability
of landslide susceptibility models. In summary, this research provides valuable insights into the
effective utilization of ML models for accurate landslide susceptibility mapping in Arunachal
Pradesh. It facilitates proactive landslide risk management and disaster mitigation efforts,
benefiting the region's stakeholders.

Keywords: Landslides, Landslide Susceptibility, Feature selection, Mutual Information,

Accuracy, ROC-AUC curve, Machine Learning.

1. Introduction

There have been worries regarding the need for the effective identification of landslide-prone
areas as a result of the increased frequency of landslides in recent years, which is causally related
to the effects of climate change. In order to assess risk and suggest mitigating measures, this
research project intends to establish a framework for modeling landslide susceptibility using
computational intelligence techniques.

Maps of landslide-prone areas have benefited greatly from the employment of statistical and
machine-learning prediction models, especially in areas with a dearth of geotechnical data.
Particularly in large-scale landslide mapping, where physically based procedures may encounter
implementation difficulties, these data-driven approaches have demonstrated encouraging
outcomes.

This research project's main goal is to examine and evaluate alternative modeling strategies for
determining landslide vulnerability. The emphasis is on creating and enhancing intelligent
models that efficiently include geographic and temporal variables for regional-scale landslide
prediction using AI and machine learning approaches. In order to address the substantial
difficulties given by the worldwide change in landslip risk prediction and mitigation, the project
emphasizes the assessment and performance monitoring of these modeling methods.

The study emphasizes the significance of employing high-quality data to provide accurate
modeling findings in order to accomplish this goal. The study investigates several computational
intelligence techniques and assesses how well they capture the intricate connections between
landslip occurrences and relevant variables. The goal of the research is to provide light on the
advantages, constraints, and applicability of various methodologies for landslide susceptibility
modeling through a detailed analysis of their performance.

In order to contextualize the research within the current body of knowledge, recent applications
and review papers on landslip modeling and the usage of computational intelligence approaches
will also be cited. This enables a thorough comprehension of the developments and trends in the
field, as well as the applicability of the approaches used for resolving the issues related to
landslip risk assessment and mitigation.

Numerous investigations into landslide susceptibility in the Himalayan region, especially the
Western Arunachal Himalayas, have been made. These studies have employed various methods
to assess landslide susceptibility and predict landslide occurrences. Here are some notable works
in this area along with the methods used for landslide susceptibility analysis:

Ghosh et al. (2015). The study utilized GIS-based statistical models, including logistic regression
and frequency ratio analysis, to assess landslide susceptibility in the Darjeeling Himalayas. The
models were developed using landslide inventory data and various terrain, geological, and land
cover [Link] et al. (2010).This research applied support vector machine (SVM) and
frequency ratio models to assess landslide susceptibility in the Sikkim Himalaya. The models
took into account several variables, including slope gradient, aspect, lithology, land use, and
proximity to rivers and [Link] et al. (2017). To determine the susceptibility of landslides
in the Bhutan Himalayas, the study used bivariate statistical models, such as the frequency ratio
and weight of evidence models. The models incorporated several factors using GIS
methodologies, including slope, aspect, lithology, land cover, and proximity to roads and
waterways. Sundriyal et al. (2012). Frequency ratio and logistic regression models were

1
employed in this study to map landslide susceptibility in the Garhwal Himalayas. The models
used data from GIS databases to include elements including slope, aspect, elevation, curvature,
lithology, and land cover. Arora et al. in 2016. In the Chenab River Basin, researchers evaluated
the efficacy of logistic regression and artificial neural network (ANN) models for mapping
landslide vulnerability. The models also took into account the distance to highways and rivers, as
well as the slope of the area, and known aspects, analyzing the curvature, lithology, and land
cover. Landslide susceptibility mapping was carried out in the Indian Darjeeling Himalayan
region as part of a study by Singh et al. (2021). The researchers used a hybrid strategy that
included logistic regression and frequency ratio models. The outcomes showed how well the
method worked at locating landslide-prone locations and gave useful information for landslide
risk management. Sharma et al. (2020) examined landslide susceptibility mapping in the Indian
state of Uttarakhand using machine learning techniques. The authors combined many techniques,
including logistic regression, support vector machines, and random forests, as part of an
ensemble modeling strategy. The study demonstrated how crucial it is to employ a range of
models in order to obtain accurate and reliable estimations of landslip vulnerability. Jaiswal et al.
(2019) carried out a landslide susceptibility investigation in the Indian Kumaon Himalayas. The
analytic hierarchy process (AHP) and weighted overlay approaches were utilized by the
researchers to find some landslide susceptibility zones. They mostly considered some crucial
factors like lithology, land covers, and slope for finding the landslide risk zones. If we look at the
study of Goyal et al. (2018), then it is one of the complete evaluations of landslide susceptibility
in the Indian Sikkim region. In this paper, the author used some machine learning models like
support vector machines and artificial neural networks. Features need to be selected carefully to
get better accuracy from these ML models. In the study by Mishra et al. (2022), the author
studied for Nilgiri district of Tamil Nadu, India where he used some popular machine learning
models like logistic regression, frequency ratio analysis, and fuzzy logic modeling. All these
models are combined to make some hybrid strategy by considering some crucial factors like
rainfall, aspects, elevation and many other features.

Due to its high topography, intense monsoon rains, seismic activity, and complicated geological
structure, the Western Arunachal Himalaya region of India experiences a lot of landslides. This
literature review, which also provides an overview of past studies on landslide modeling and
prediction in the Himalayan region with an emphasis on the Western Arunachal Himalaya,
focuses on the study of "Modelling and Predicting Landslides in Western Arunachal Himalaya,
India". It is significant to note that based on the data availability, study region features, and
research aims, the specific methods employed for landslide susceptibility analysis may change.
The Himalayan region's landslide susceptibility modeling and prediction can be improved, and

2
there are chances to better understand landslide dynamics in these difficult terrains, thanks to
ongoing developments in geospatial technologies and machine learning algorithms.

1.1 Study Area

Arunachal Pradesh is the easternmost area of the Himalayan mountain range. The latitude of
Arunachal Pradesh is 91°30′ E to 96°E and the latitude of Arunachal Pradesh is 26°28′ N to
29°30′ N and it contains the eastern Himalayan pattern. The border of Arunachal Pradesh shares
international borders. Arunachal borders are the states of Assam and Nagaland and The
international boundaries of Arunachal Pradesh are Bhutan in the west, and Myanmar in the east.
The study area of landslides would typically involve identifying, analyzing, and assessing
landslide-prone regions within the state. The geographical conditions of Arunachal Pradesh have
a significant impact on its climatic characteristics. Additionally, this region falls under Zone V,
the highest seismic zone, owing to its geographical location and active tectonic activities. Some
of the districts of Arunachal Pradesh have more susceptibility to landslides. These districts are
Tawang District, West Kameng District, East Kameng District, Papum Pare District , Lower
Subansiri District . Arunachal Pradesh receives heavy rainfall throughout the year, particularly
during the monsoon season. The annual rainfall of this state received 150cm-200cm.

3
 Fig 1.1: Study Area of Arunachal Pradesh

Fig1.2: Landslide or non-landslide zone of Arunachal Pradesh

3. Proposed methodology:

4
 Fig3.1 : Methodology

3.1 Data collection:

Data collection: In order to predict and forecast landslides, it is necessary to develop a
comprehensive landslide inventory. Landslide databases play a vital role in recording the
geographical location and characteristics of past landslides that have occurred in the area .so
firstly we got the landslide point of our study area ([Link]
landslide-catalog ). To develop a noble and accurate landslide susceptibility model, it is crucial to
carefully select appropriate conditioning factors. In this research, a total of 9 conditioning factors
were selected on the basis of previous literature
and related studies about Arunachal Pradesh. The selected factors are DEM, Slope, Aspects, Soil,
NDVI, NDWI, LULC, Distance from epi-center, and Geology. These factors were deemed
significant based on their influence on landslide occurrences and their relevance to the study area.
The description of each landslides conditioning factor is given below:

5
 Fig 3.2: Data Collection

1. Dem: DEM(Digital Elevation Model) is widely used in landslide data analysis and
studies due to its critical role in understanding the terrain characteristics and topographic
influences on landslides. It is used also for the precise prediction of landslides
susceptibility. Through DEM data we have to get data on slope, aspect, plain curvature,
and so on. The maximum elevation reaches 6516m, and the lowest elevation reaches 90 m
in the study area.
2. Slope: slope refers to the steepness or inclination of the terrain in a given area. It plays a
crucial role in assessing the potential for landslides' occurrence. The slope is an important
factor because steep slopes are generally more susceptible to landslides due to the
increased gravitational forces acting on the materials.
3. Aspect: Aspects refer to the compass direction that a slope faces. By considering the
aspect of slopes, researchers can better understand the spatial variability of environmental
conditions that affect slope stability. It provides information about the orientation of a
slope or terrain.

6
4. Soil: Soil is a fundamental parameter in landslide susceptibility studies on the grounds
that the properties and qualities of the soil assume a huge part in slope stability. Shear
strength, immersion, pore pressure, soil type and surface, slant point, and soil properties
assume a significant part in assessing landslide susceptibility. This data supports figuring
out the strength of slope, recognizing regions with higher landslides susceptibility, and
carrying out fitting measures for landslide prevention and mitigation.
5. NDVI: Normalized Difference Vegetation Index (NDVI) is often considered a parameter
in landslide susceptibility studies because it provides information about vegetation cover
and health, which can be indicators of slope stability. NDVI is derived from satellite
imagery and measures the difference in the reflectance of near-infrared (NIR) and visible
red (RED) light wavelengths.
6. NDWI: NDWI is a remote sensing index that quantifies the presence of water or
moisture content in vegetation or soil. While NDWI can be useful for mapping water
bodies and detecting areas prone to flooding, its direct application in landslide
susceptibility analysis is limited. Landslides susceptibility studies typically focus on
factors related to slope stability, geological conditions, terrain characteristic, land cover,
and rainfall patterns, among others. These factors help assess the potential for landslides
to occur.
7. LULC: Land Use/Land Cover (LULC) data categorizes and maps different land cover
types such as forests, agriculture, urban areas, bare soil, and water bodies. Surface
Roughness and soil erosion, Vegetation cover and root strength Human activities and
engineering structure, land cover patterns, and hydrological conditions are important in
assessing landslides susceptibility.
8. Distance from epi-center: It helps in identifying areas with higher susceptibility to
landslides due to their proximity to the epi-center, understanding the amplification effects
of local geology, evaluating the proximity to active faults, and assessing the potential
secondary effects of earthquakes.
9. Geology: By considering geology as a parameter in landslides studies, researchers can
assess the strength and mechanical properties of rocks and soils, evaluate the influence of
geological structures, identify areas prone to weathering and erosion, understand
groundwater conditions, and anticipate the types of slope processes that may occur.

7
 Fig 3.3: DEM, Slope, Aspect, LULC , NDVI or NDWI

3.2 Data preprocessing:

Preprocessing is done to the data after it has been collected. This entails cleansing the data,
handling outliers, and dealing with missing values. In the future, it may also be possible to merge
many datasets to produce a comprehensive dataset for study. This time, though, the team
processed the data manually without the use of a method or piece of code. Before conducting
additional analysis and modeling, these procedures were taken to make sure the data's quality and
integrity were maintained.

8
3.3 Feature selection:
In order to determine the factors that have the greatest influence on landslide vulnerability,
feature selection procedures are used. Exploratory data analysis is done to learn more about the
traits and connections of the variables that were gathered. The characteristics that have the
greatest influence on the occurrence of landslides are identified using statistical techniques.
The mutual information strategy was used in this study to choose features. Mutual information
calculates how statistically dependent two variables are and estimates how much knowledge each
variable has of the other. In this instance, we determined the mutual information between each
feature and the target variable, which reflects the degree of relevance or knowledge each
characteristic imparts regarding the target variable.
Ranking the features according to the mutual information scores was a step in the feature
selection process. Higher mutual information scores for features show a potential relevance for
our research because they suggest a stronger association or reliance with the target variable.
We established a threshold value (the mutual information score of a feature should not be equal
to 0) to identify the most informative features. Only features with mutual information scores over
the cutoff were chosen for additional examination. We reduced the dataset's dimensionality by
employing a threshold to guarantee that only the most important attributes were taken into
account, enhancing the effectiveness and understandability of our models.
Following that, the chosen features were employed as input variables for modeling and analysis
tasks. We sought to find the most relevant and informative features that may make a major
contribution to our investigation of landslide susceptibility by using the mutual information-
based feature selection approach.
Overall, we were able to find a subset of features that are extremely pertinent to our research of
the susceptibility of landslides through the use of mutual information-based feature selection,
improving the precision and interpretability of our models while lowering computing complexity.

3.4 Model selection:

Based on the nature of the data and research objectives, suitable AI models are selected for

9
landslide susceptibility mapping. This may involve choosing machine learning algorithms or
deep learning architectures. The selection is based on the performance, interpretability, and
scalability of the models.
By considering various factors, XGBoost, Random Forest, SVM, RBFN, and Logistic Regression
were chosen as they provide a diverse set of models that can address the requirements of the
landslide susceptibility analysis(XGBoost, Random Forest, and Logistic Regression are
computationally efficient models, allowing for scalability to large datasets.), demonstrate
performance in similar tasks, offer interpretability(Logistic Regression stands out among the
selected models for its interpretability. The coefficients of the logistic regression model can
provide insights into the relative importance and direction of influence of the input features on
the predicted landslide susceptibility.), and are suitable for handling the characteristics of the
dataset at hand.

3.5 Model training:

The selected models are trained using preprocessed data. The dataset is split into training and
validation sets. The models are trained using the training set, and their parameters are optimized
through techniques such as cross-validation and hyperparameter tuning. The goal is to achieve
the best possible performance of the models.

3.6 Model evaluation:

Once the models are trained, they are evaluated using the validation set. The model evaluation is
conducted using the ROC-AUC curve and accuracy as the performance metrics. Here's an
explanation of how these metrics are used for evaluating the models:

3.6.1 ROC-AUC Curve:

● The ROC curve is a visual representation of a classification model's performance at
different classification thresholds.
● AUC (Area Under the Curve) is a single numeric value that summarizes the overall model
performance.
● A higher AUC value indicates better model performance in distinguishing between
classes.

10
3.6.2 Accuracy:
● Accuracy measures the proportion of correct predictions out of the total predictions made
by the model.
● It is a simple metric but may not capture the full picture of model performance, especially
in imbalanced classes or when certain misclassifications are more critical.
● Using accuracy alongside other metrics like the ROC-AUC curve provides a more
comprehensive evaluation of the model.
The evaluation helps determine the models' effectiveness in predicting landslide susceptibility.

3.7 Model refinement:

Based on the outcomes of the evaluation, the models' accuracy and resilience are further
improved during the model refining step. This phase includes a number of tasks, such as
modifying the model's parameters, investigating alternative methods, and adding new features. It
is significant to highlight that the process of refining and the related duties are taken into account
for future work.

3.7.1 Adjusting Model Parameters:

The parameters of the models will be adjusted to maximize their effectiveness. The best results
will be chosen after exploring various parameter value combinations using methods like grid
search and random search.

3.7.2 Exploring Different Algorithms:

We will investigate alternative machine learning techniques to see if they may enhance the
performance of the model. Trying out algorithms like RNN and CNN deep learning algorithms,
LDA, QDA, etc or ensemble techniques like bagging or boosting may be necessary to achieve
this.

3.7.3 Incorporating Additional Features:

We'll look at whether it's possible to add more pertinent features to the models. These
characteristics might offer further data or boost the models' capacity for forecasting. To create
new features from existing ones, feature engineering approaches like developing interaction

11
terms or polynomial features may be used.
Iterative and experimental in nature, the refinement process aims to improve the models'
accuracy and robustness.
Future research will make the aforementioned corrections and improvements to further improve
the models' ability to forecast landslip vulnerability.

3.8 Technology
1. GEE (Google Earth Engine ): Google Earth Engine (GEE) is a powerful platform that
offers various benefits for landslide analysis and research. We use GEE for :
a. Access to Diverse Data: GEE provides access to a vast collection of geospatial
data, including satellite imagery, climate data, terrain data, and more. This rich
data repository allows researchers to access and analyze relevant datasets for
landslide studies. GEE's data catalog includes data from multiple sources, such as
Landsat, Sentinel, MODIS, and other remote sensing platforms, which are
valuable for monitoring and assessing the landslide-prone area
b. Integration of Earth Observation Data: GEE considers the joining of various sorts
of geospatial information, empowering specialists to consolidate information from
different sources and perform multi-layer analysis. This joining works with the
appraisal of different elements impacting landslide susceptibility , for example,
geology, land cover, precipitation designs, topographical data, and that's just the
more. By incorporating different datasets, scientists can produce complete
landslide susceptibility models.
c. Code Sharing and Collaboration: GEE promotes code sharing and collaboration
among researchers. We can develop custom scripts and algorithms in the GEE
Code Editor and share them with the community.
d. Visualization and Communication: GEE offers strong perception capacities,
permitting analysts to make interactive maps , time series animations, and other
visual representation of their information and analysis results.

2. ArcGIS Pro: ArcGIS Pro is a widely used software tool for landslide susceptibility analysis
due to its robust geospatial analysis capabilities and comprehensive suite of tools. We use

12
ArcGIS Pro for Landslides Susceptibility are :

a. Geospatial Analysis and Modeling: ArcGIS Pro offers a set-up of strong geospatial
analysis instruments that are fundamental for landslides susceptibility studies. We can
perform terrain analysis, calculate slope and aspect, generate terrain derivatives, Plain
curvature, Profile Curvature and derive geomorphological parameters. The software also
provides tools for spatial interpolation, and spatial overlay operations.
b. Susceptibility Modeling and Mapping: we use it for mapping and it supports the
development and implementation of various landslides susceptibility modeling
techniques. We can create high-quality maps. We can create compelling visual
representations of landslide susceptibility maps, overlay multiple layers, and generate
informative reports to support decision-making.

3. Machine Learning: The foundation of our project is machine learning techniques. XGBoost,
Random Forest, SVM, RBFN, and LR are some of the machine learning (ML) methods we use to
build prediction models for landslide susceptibility and mitigation mapping.

4. Methods for Feature Selection: To determine the most important traits that affect landslide
susceptibility, we use selective methods like Mutual Information. This aids in lowering the
dataset's dimensionality and choosing the most useful variables for model training.

5. Analytical Methods: To assess the effectiveness of our ML models, accuracy and the AUC-
ROC curve are the analytical methods used. The AUC-ROC curve offers information about the
models' capacity to differentiate between several classes, while accuracy assesses how accurate
the predictions are overall.

6. Programming Languages and Libraries: To develop the ML models, feature selection

procedures, and performance evaluation strategies, we make use of programming languages like
Python and R as well as scikit-learn, XGBoost, and pandas.

7. Data Visualisation Tools: To produce visual representations of the data, such as ROC curves,
feature significance plots, and other educational visualizations, data visualization tools like
Matplotlib and Seaborn are used.

13
4. Result and Analysis

1.) Accuracy: For the job of mapping landslip vulnerability and mitigation, the accuracy
ratings of the five machine learning models are as follows:
○ XGBoost: 0.58
○ SVM: 0.64
○ Random Forest: 0.58
○ RBFN: 0.70
○ LR: 0.76
These scores indicate the proportion of correctly classified instances by each model,
providing an assessment of their predictive performance.

2.) ROC Curve and AUC:

The ROC curves visually represent the performance of the models by plotting the true
positive rate against the false positive rate at various classification thresholds. The Area
Under the Curve (AUC) scores quantitatively measure the models' ability to distinguish
between landslide and non-landslide zones.

14
Fig4.1: Comparison Between Random Forest, XGBoost, and SVM Models.

15
 Fig 4.2: Comparison Between RBFN and Logistic Regression Models.

3.) Feature Selection:

To identify the most relevant features for the analysis, we employed mutual information-
based feature selection. This process involved ranking the features based on their mutual
information scores, which indicate the strength of the relationship between each feature and the
target variable.

Based on the mutual information scores, a subset of features was selected for the analysis. We
carefully considered a range of 7 features, striking a balance between information content and
model complexity.

16
Fig 4.3: Features Ranked by their Mutual Information Scores

We learned more about the effectiveness and applicability of the machine learning models in the
context of mapping landslide susceptibility and mitigation by merging the accuracy scores, ROC
curves with AUC scores, and the chosen features through mutual knowledge.

5. Conclusions and Future Scope

For landslide susceptibility mapping in Arunachal Pradesh, this study greatly advances
knowledge and implementation of machine learning (ML) models. Policymakers, researchers,
and practitioners can use the data to build focused mitigation initiatives, create useful land-use
plans, and make decisions that will lessen the impact of landslides in the area.
The integration of different ML models, such as logistic regression, frequency ratio, support
vector machines, and artificial neural networks, along with Geographic Information System
(GIS) methodology has shown how important it is to take into account a variety of factors,
including topographical features, the lithology of the land, the slope characteristics, and the
proximity to roads and rivers. Accurate identification and categorization of landslide-prone
locations have been obtained by including these characteristics in the models, permitting efficient
landslide management and risk reduction techniques.
However, it is crucial to recognise the difficulties that still remain in this area, such as the
scarcity of data, the lack of assurance around the model inputs and parameters, and the
requirement to take the dynamics of climate change into account. To improve data collection
efforts, create models with exact input parameters, and evaluate the potential effects of climate
change on the occurrence of landslides, more study is necessary. These initiatives will help to

17
increase the Western Arunachal Himalayas' and other landslide-prone areas' landslide
susceptibility estimates' accuracy and dependability. Future research could study additional
evaluation measures like Kappa and F1-score to give a more thorough knowledge of model
performance. Additionally, the utilization of ensemble learning methods such as bagging and
boosting may further enhance the accuracy and robustness of the models.
By addressing these aspects, future studies can advance the comprehension and application of
landslide susceptibility assessment. The continuous improvement of ML models and the
integration of additional techniques will contribute to enhancing the precision, reliability, and
effectiveness of landslide susceptibility mapping, ultimately aiding in the development of
proactive measures for landslide prevention and mitigation.

Future scope :
Future research and improvements in landslide risk estimation and mitigation measures can
contribute to enhancing our understanding of landslides and developing more effective strategies
to mitigate their impact. Here are some potential areas of focus for future scope are :

1. Improved Data Acquisition and Integration: Further advancement in remote detecting

technologies, for example, higher-goal satellite imagery, LiDAR information, and
aerial photogrammetry, can improve the quality and detail of info information for
landslides risk assessment. Combination of different datasets, including geographical,
hydrological, and environment information, can give a more far reaching
comprehension of landslide- inclined regions.

2. Development of Advanced Models: Exploration can focus on the improvement of

advanced models that consolidate numerous variables affecting landslides, for
example, territory qualities, land cover, precipitation designs, soil properties, and
human exercises. integration of machine learning algorithms, probabilistic
demonstrating, and ensemble procedures can lead to more exact and powerful
landslides risk models.

18
 3. Refinement of Model Parameters: Future studies can focus on adjusting the parameters
of the models to maximize their effectiveness. Techniques like grid search and
random search can be employed to explore various parameter value combinations and
select the best-performing models.

4. Early Warning Systems: Improving on early advance notice frameworks for

landslides can essentially decrease the expected death and property. Future research
can focus on the development of strong checking frameworks that coordinate constant
information from different sources, for example, precipitation measures, ground
mishappening sensors, and satellite-based perceptions. Integration with advanced
information examination and prescient demonstrating methods can improve the
accuracy and practicality of landslide early advance warning systems.

References

1. Achour, Y., & Pourghasemi, H. R. (2020). How do machine learning techniques help in
increasing the accuracy of landslide susceptibility maps? Geoscience Frontiers, 11(3),
871–883. [Link]
2. Chen, X., & Chen, W. (2021). GIS-based landslide susceptibility assessment using
optimized hybrid machine learning methods. Catena, 196(July 2020), 104833.
[Link]
3. Goetz, J. N., Brenning, A., Petschko, H., & Leopold, P. (2015). Evaluating machine
learning and statistical prediction techniques for landslide susceptibility modeling.
Computers and Geosciences, 81, 1–11. [Link]
4. Kainthura, P., & Sharma, N. (2022). Hybrid machine learning approach for landslide
prediction, Uttarakhand, India. Scientific Reports, 12(1), 1–23.
[Link]
5. Kavzoglu, T., Colkesen, I., & Sahin, E. K. (2019). Machine learning techniques in
landslide susceptibility mapping: A survey and a case study. Advances in Natural and
Technological Hazards Research, 50, 283–301. [Link]
3_13
6. Ma, Z., Mei, G., & Piccialli, F. (2021). Machine learning for landslide prevention: a

19
 survey. Neural Computing and Applications, 33(17), 10881–10907.
[Link]
7. Meena, V., Kumari, S., & Shankar, V. (2022). Physically based modelling techniques for
landslide susceptibility analysis: A comparison. IOP Conference Series: Earth and
Environmental Science, 1032(1). [Link]
8. Ngo, T. Q., Dam, N. D., Al-Ansari, N., Amiri, M., Phong, T. Van, Prakash, I., Le, H.
Van, Nguyen, H. B. T., & Pham, B. T. (2021). Landslide Susceptibility Mapping Using
Single Machine Learning Models: A Case Study from Pithoragarh District, India.
Advances in Civil Engineering, 2021. [Link]
9. Nhu, V. H., Mohammadi, A., Shahabi, H., Ahmad, B. Bin, Al-Ansari, N., Shirzadi, A.,
Clague, J. J., Jaafari, A., Chen, W., & Nguyen, H. (2020). Landslide susceptibility
mapping using machine learning algorithms and remote sensing data in a tropical
environment. International Journal of Environmental Research and Public Health,
17(14), 1–23. [Link]
10. Patra, P., & Devi, R. (2015). Assessment, prevention and mitigation of landslide hazard in
the Lesser Himalaya of Himachal Pradesh. Environmental & Socio-Economic Studies,
3(3), 1–11. [Link]
11. Saha, S., Bera, B., Shit, P. K., Sengupta, D., Bhattacharjee, S., Sengupta, N., Majumdar,
P., & Adhikary, P. P. (2023). Modeling and predicting landslides in Western Arunachal
Himalaya, India. Geosystems and Geoenvironment, 2(2), 100158.
[Link]
12. Youssef, A. M., & Pourghasemi, H. R. (2021). Landslide susceptibility mapping using
machine learning algorithms and comparison of their performance at Abha Basin, Asir
Region, Saudi Arabia. Geoscience Frontiers, 12(2), 639–655.
[Link]

Source Code:

20
[Link]

21