Review on the Application of Machine Learning Methods

in Landslide Susceptibility Mapping

Deborah Simon Mwakapesa

, Ye Li

, Wang Xiangtai

, Guo Binbin

and Mao Yimin

School of Civil and Surveying Engineering, Jiangxi University of Science and Technology, Ganzhou, China

School of Information Engineering, Jiangxi University of Science and Technology, Ganzhou, China

Keywords: Machine Learning, Supervised Learning, Unsupervised Learning, Deep Learning, Landslide, Landslide

Susceptibility Mapping

Abstract: Machine learning is a very important in computer science field which has gained attention in numerous

applications. This paper reviewed various machine learning methods including supervised and unsupervised

learning and highlighted their applications, advantages and disadvantages in landslide susceptibility

mapping. The review has also mentioned the challenges of machine learning algorithms for achieving

higher performance accuracy from the supervised and unsupervised learning algorithms during landslide

susceptibility. Moreover, highlights on the application of deep learning methods as the current research in

landslide susceptibility mapping has also been reported. Finally, this paper argued the necessity of thorough

preparation of relevant and enough data being significant important to obtain high performance results from

the review methods.

1 INTRODUCTION

Landslides (Cruden, 1991) involve the downward

movement of various earth materials such as soil,

mass or rocks, debris, and others, as a result of

gravity. Population growth, settlement and economic

development, as well as climatic changes are learnt

to be the major triggers of landslides. These

landslides have brought up highly variable impacts

on both physical as well as human environment. It

has been recorded that not less than 3.5 million

kilometers square of the total land area in the world

have been affected and are still susceptible to

landslides (Dilley et al., 2005). Also, according to

World Health Organization (WHO), between 1998

and 2017 approximately 4.8 million people were

affected by landslides which also lead to more than

18,000 deaths. It is named among the very

dangerous and disturbing disasters in the world.

Thus, investigating and identifying areas susceptible

to landslides for taking control as well as preventing

measures is very important. One of the common

measure is landslide susceptibility mapping (LSM,

conducted on different landslide influencing factors

such as geological, geomorphological and

hydrological factors) using various methods such as

Machine Learning (ML) which has gained much

attention among researchers from different places in

the world (Wang et al., 2020; Hu et al., 2020; Mao et

al., 2021a; Mao et al., 2021b).

In the following sections, the machine learning

methods are described as well as their applications

in LSM are briefly highlighted.

2 REVIEW OF MACHINE

LEARNING AND ITS

APPLICATION IN LSM

2.1 Machine Learning (ML)

ML (Mitchell Tom, 1997) involves computer

algorithms which improves through experience and

by the use of data. ML algorithms construct models

by learning from data and self-improve. These

algorithms are applied in different applications

including computer vision, medicine, speech

recognition, and disaster prediction. ML is divided

into four types: supervised learning, semi supervised

learning, unsupervised learning and reinforcement

learning. In LSM, the supervised and unsupervised

Mwakapesa, D., Li, Y., Xiangtai, W., Binbin, G. and Yimin, M.

Review on the Application of Machine Learning Methods in Landslide Susceptibility Mapping.

DOI: 10.5220/0010790400003167

In Proceedings of the 1st International Conference on Innovation in Computer and Information Science (ICICIS 2021), pages 69-72

ISBN: 978-989-758-577-7

learning are commonly used (Buhmann, 1992;

Boussemart et al., 2011).

2.1.1 Supervised Learning (SL)

In SL (Sathya and Annammna, 2013), methods such

as classification, regression and prediction are

trained using labeled examples, such as an input

where the desired output (labels) is known. For

example, having dataset labeled either landslide or

non-landslide. The algorithm receives a set of inputs

samples along with the corresponding label; the

algorithm learns by comparing the actual output with

the corresponding label to find errors and then

adjusts the model accordingly. When there is an

additional unlabeled data, the SL methods use

patterns to predict the values of the label. Thus, in

LSM, SL is applied in predicting future landslide

events. Some SL algorithms that have commonly

applied in LSM in recent years include support

vector machine (SVM, Yu and Lu, 2018; Anik and

Suli, 2020), logistic regression (Feby et al., 2020;

Paul and Alejandra, 2021), classification and

regression trees (Chen et al., 2017; Sun et al., 2021),

decision trees (Mao et al., 2017; Kavzoglu et al.,

2019; Guo et al., 2021) random forest (Chen et al.,

2017; Sun et al., 2021), weight of evidence (Anik

and Suli, 2020) and artificial neural network

(Bragagnolo et al., 2020; Lucchese et al., 2021).

2.1.2 Unsupervised Learning (USL)

USL (Hinton and Sejnowski, 1999) such as

clustering algorithms are used on data that has no

labels, so the algorithm must determine where the

data belong to with the aim of exploring the data and

find its patterns. Up to date, there are very few USL

algorithms that have been proposed in LSM,

including k-means (Wang et al., 2017, Guo et al.,

2021), Fuzzy C-means (FCM) and K-means particle

swam optimization (KPSO, Wan Shiuan 2013; Wan

Shiuan 2015), k-means and Hiearchical clustering

(Pokharel et al., 2020), CA-AQD (Hu et al., 2020),

AHC-OLID (Mao et al., 2021a), and OA-HD (Mao

et al., 2021b). From the analysis of the current

proposed USL methods in LSM, are hybrid methods,

which are the modification of the traditional USL

methods, while the traditional USL methods have

not been explored in length as compared to the SL

methods.

2.2 Discussion

In this paper, the study was conducted on the

application of ML methods in landslide

susceptibility mapping, on the basis of the

developments that have been proposed and reported

by researchers. The major application of ML

methods in LSM can be observed in the area of SL

algorithms as most of the studies published in

various journals are based on SL. This is due to the

advantages possessed by the SL methods, including:

SL allows researchers to collect of produced data

based on experience; this experience enhances

performance criteria optimization; and SL gives an

exact idea about the classes in the data such as

landslide and non-landslide classes. Thus, these

advantages make it easy to implement the SL

methods in LSM. However, their applications are

limited in various ways: inability to discover deep or

unknown patterns in the data, thus, the results may

not always be accurate; the accuracy of the methods

depends on the available data, they require a lot of

samples from the labels or classes for training to

obtain high accuracy, whereby, in real situations, it

is not easy to obtain landslide data especially when

dealing with large study areas; also, the involved

training process consumes a lot of computation time

especially with large datasets from large study areas.

On the other side, the current proposed USL

methods in LSM, have shown some advantages over

the SL methods including: with USL, the methods

learn and discover the features or patterns present in

the data, then finds the similarities and

dissimilarities in the data which make it easy to

group them into different groups (classes) in absence

of the data labels; discovering of features in the data

make it easy to process the data even when other

unlabeled data are added; also, this process does not

consume a lot of computation time. Despite of their

advantages they also have disadvantages, such as the

in some situations, their results may not be very

accurate as there is no training of data during the

process in some cases, human intervention might be

needed to validate the results; in LSM projects with

real data, the USL involves feeding of data to the

algorithm continuously which may result in

inaccurate results as well as time consuming; also,

when there are a lot of features in the data the

process becomes complex.

However, from the above analysis on both cases,

the performance of these methods depend on the

available data. Thus, thorough and careful

preparation of the data is a very significant stage in

LSM while using these methods. Also, it has been

observed that in both cases of ML methods, their

ability to learn deep features from the data is very

shallow, as they have one hidden layer or none.

Thus, their performance results may not be very

ICICIS 2021 - International Conference on Innovations in Computer and Information Science

accurate when a data with deep and complex

features is involved.

Furthermore, the LSM literature shows that

currently, the research is directing to proposing LSM

models based on deep learning methods

(Goodfellow et al., 2016) which tend to have better

features as compared to the former SL and USL

proposed methods (Nhu et al., 2020). This is because

the DL methods possess hidden layers or deep

structures which facilitate the learning of deep and

complex features in the data, thus the name Deep

Learning. They also make it easy to process big

datasets from larger study areas. So far, there are

very studies that have been proposed and have so far

shown better performance results compared to the

prevailing methods. Some of the LSM deep learning

models that have been proposed so far includes:

deep neural networks (DNN, Kanu et al., 2021;

Dong et al., 2020; Bui et al., 2020; Nhu et al., 2020;

Dou et al., 2020); convolutional neural networks

(Bui et al., 2020; Dou et al., 2020; Nhu et al., 2020;

Yi et al., 2020; Fang et al., 2020; Ngo et al., 2021;

Bragagnolo et al., 2021). And so far, these methods

have shown promising performance results in their

implementations. However, they also have some

limitations, such as the fact that the DL models

requires a lot of samples to train the models, and in

cases where it is not easy to obtain many samples,

the DL performance becomes limited.

3 CONCLUSIONS

This study was reviews the application of machine

learning methods, the supervised and unsupervised

learning, in landslide susceptibility mapping. The

two types have been briefly discussed, they

advantages and disadvantages have also been

provided. At last, we also looked at the deep

learning method which as per the literature review it

has shown to perform better than the machine

learning methods. This learning methodology has

great significance. Although it have not been

explored much as compared to machine learning, it

can be very helpful in research. It has also been

observed that, the performance of all the reviewed

methods depends on the data. Therefore, the

selection and preparation of relevant and enough

data is crucial for the methods to work efficiently,

especially with the deep learning. Moreover, this

paper should also contribute to the collection of

various machine learning application in LSM for

easy reference.

ACKNOWLEDGEMENTS

This study was supported by the National Key

Research and Development program

(2018YDC1504705) and the National Natural

Science Foundation of China (41562019).

REFERENCES

Cruden, M, 1991. A simple definition of a landslide.

Bulletin of the International Association of

Engineering Geology.

Buhmann, J, Kuhnel, H, 1992. Unsupervised and

supervised data clustering with competitive neural

networks. [Proceedings 1992] IJCNN International

Joint Conference on Neural Networks.

Mitchell, T, 1997. Machine Learning. New York:

McGraw Hill

Hinton, G, Sejnowski, T, 1999. Unsupervised Learning:

Foundations of Neural Computation. MIT Press.

Dilley, M, Chen, S, Deichmann, U, Lerner, L, Arthur, L,

Arnold, M, 2005. Natural Disaster Hotspots: A Global

Risk Analysis. Washington, DC: World Bank.

Boussemart, Y, Mary, L, Cummings, Jonathan, F,

Nicholas R, 2011. Supervised vs Unsupervised

Learning for Operator State Modeling in Unmanned

Vehicle Settings. Journal of Aerospace Computing,

Information, and Communication.

Sathya, R, Annamma A, 2013. Comparison of Supervised

and Unsupervised Learning Algorithms for Pattern

Classification. International Journal of Advanced

Research in Artificial Intelligence.

Wan, Shiuan, 2013. Entropy-based particle swarm

optimization with clustering analysis on landslide

susceptibility mapping. Environmental Earth Science.

Wan Shiuan, 2015. Construction of knowledge-based

spatial decision support for landslide mapping using

fuzzy clustering and KPSO analysis. Arabian Journal

of Geoscience.

Goodfellow, I, Bengio, Y, Courville, A, 2016. Deep

learning. MIT Press.

Chen W, Xiaoshen X, Jiale W, Biswajeet P, Haoyuan H,

Dieu B, Zhao D, Jianquan M, 2017. A comparative

study of logistic model tree, random forest, and

classification and regression tree models for spatial

prediction of landslide susceptibility, CATENA.

Wang Q, Wang Y, Niu Q, 2017. Integration of

information theory, k-means cluster analysis and the

logistic regression model for landslide susceptibility

mapping in the three gorges area, China. Remote

Sensing.

Kavzoglu T, Colkesen I, Sahin EK, 2018. Machine

learning techniques in landslide susceptibility

mapping: a survey and a case study. Advances in

Natural and Technological Hazards Research.

Yu H, Lu Z, 2018. Review on landslide susceptibility

mapping using support vector machines, CATENA.

Review on the Application of Machine Learning Methods in Landslide Susceptibility Mapping

Anik Saha, Sunil Saha, 2020. Comparing the efficiency of

weight of evidence, support vector machine and their

ensemble approaches in landslide susceptibility

modelling: A study on Kurseong region of Darjeeling

Himalaya, India. Remote Sensing Applications:

Society and Environment.

Bo, Yu, Fang Chen, Chong Xu, 2020. Landslide detection

based on contour-based deep learning framework in

case of national scale of Nepal in 2015. Computers &

Geosciences.

Bragagnolo, L, R.V. da Silva, Grzybowski, J 2020.

Landslide susceptibility mapping with r.landslide: A

free open-source GIS-integrated tool based on

Artificial Neural Networks. Environmental Modelling

& Software.

Dong, D, Abolfazl, J, Mahmoud, B, Davood, M,

Chongchong, Q, Hossein, M, Tran, P, Hai-Bang, L,

Tien-Thinh, L, Phan, T, Chinh, L, Nguyen, Q, Bui, T,

Binh Thai, P, 2020. A spatially explicit deep learning

neural network model for the prediction of landslide

susceptibility. CATENA.

Bui, T, Paraskevas, T, Viet-Tien N, Ngo, L, Phan, T,

2020. Comparing the prediction performance of a

Deep Learning Neural Network model with

conventional machine learning models in landslide

susceptibility assessment. CATENA.

Feby B, Achu, L, Jimnisha, K, Ayisha A, Rajesh, R, 2020.

Landslide susceptibility modelling using integrated

evidential belief function based logistic regression method:

A study from Southern Western Ghats, India. Remote

Sensing Applications: Society and Environment.

Hu, J, Xu, K, Wang, G. et al., 2020. A novel landslide

susceptibility mapping portrayed by OA-HD and K-

medoids clustering algorithms. Bulletin Engineering

Geology and the Environment.

Dou J, Ali P. Y, Abdelaziz M, Ataollah S, Hoang N,

Yawar H, Ram A, Yulong C, Binh Thai P, Hiromitsu

Y, 2020. Different sampling strategies for predicting

landslide susceptibilities are deemed less

consequential with deep learning. Science of the Total

Environment.

Pokharel B, Omar A, Ali A, et al., 2020. Spatial clustering

and modelling for landslide susceptibility mapping in

the north of the Kathmandu Valley, Nepal. Landslides.

Nhu V, Nhat-Duc H, Hieu N, Phuong Thao N, Tinh Thanh

B, Pham H, Pijush S, Dieu Tien B, 2020.

Effectiveness assessment of Keras based deep learning

with different robust optimization algorithms for

shallow landslide susceptibility mapping at tropical

area, CATENA.

Yi Y, Zhijie Z, Wanchang Z, Huihui J, Jianqiang Z, 2020.

Landslide susceptibility mapping using multiscale

sampling strategy and convolutional neural network:

A case study in Jiuzhaigou region. CATENA.

Fang Z, Yi W, Ling P, Haoyuan H, 2020. Integration of

convolutional neural network and conventional

machine learning classifiers for landslide susceptibility

mapping. Computers and Geosciences.

Bragagnolo L; L.R, Rezende, R.V. da Silva, Grzybowski

J.M.V, 2021. Convolutional neural networks applied

to semantic segmentation of landslide scars. CATENA.

Guo Z, Yu S, Faming H, Xuanmei F, Jinsong H, 2021.

Landslide susceptibility zonation method based on

C5.0 decision tree and K-means cluster algorithms to

improve the efficiency of risk management.

Geoscience Frontiers.

Kanu M, Sunil S, Sujit M, 2021. Applying deep learning

and benchmark machine learning algorithms for

landslide susceptibility modelling in Rorachu river

basin of Sikkim Himalaya, India. Geoscience

Frontiers.

Lucchese L, Guilherme Garcia de Oliveira, Olavo Correa

P, 2021. Investigation of the influence of

nonoccurrence sampling on landslide susceptibility

assessment using Artificial Neural Networks.

CATENA.

Mao Y, Deborah SM, Wang G, Yaser AN, Zhang M,

2021a. Landslide susceptibility modelling based on

AHC-OLID clustering algorithm. Advances in Space

Research.

Mao Y, Yican L, Deborah SM, Wang G, Yaser AN,

Muhammad AK, Zhang M, 2021b. Innovative

Landslide Susceptibility Mapping Portrayed by CA-

AQD and K-Means Clustering Algorithms. Advances

in Civil Engineering.

Ngo Thao T, Mahdi P, Khabat K, Omid G, Narges K,

Artemi C, Saro L, 2021. Evaluation of deep learning

algorithms for national scale landslide susceptibility

mapping of Iran. Geoscience Frontiers.

Paul Goyes-P, Alejandra Hernandez-R, 2021. Landslide

susceptibility index based on the integration of logistic

regression and weights of evidence: A case study in

Popayan, Colombia. Engineering Geology.

Sun D, Jiahui Xu, Haijia W, Danzhou W, 2021.

Assessment of landslide susceptibility mapping based

on Bayesian hyper-parameter optimization: A

comparison between logistic regression and random

forest. Engineering Geology.

ICICIS 2021 - International Conference on Innovations in Computer and Information Science