Predicting Landslides with Machine Learning: A Data-Driven Approach

Predicting Landslides with Machine Learning: A Data-Driven Approach | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Predicting Landslides with Machine Learning: A Data-Driven Approach Karan Sarawagi, Navjot Singh, Khushwant Virdi This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-4632694/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Landslides are natural disasters that can cause significant damage to the environment and pose a serious threat to human lives and infrastructure. Early detection and identification of potential landslide-prone areas are crucial for disaster mitigation and preparedness efforts. This abstract out- lines a comprehensive approach to landslide identification uti- lizing machine learning techniques. In recent years, machine learning has emerged as a powerful tool for analyzing geospatial data and predicting geological hazards such as landslides. This research leverages a diverse range of data sources, including remote sensing imagery, topographical maps, rainfall records, and geological data, to develop a robust landslide identification model. The key components of the proposed methodology involve data preprocessing, feature engineering, and the application of various machine learning algorithms. Remote sensing data, such as satellite imagery and LiDAR data, are used to extract valuable terrain features and land cover information. Rainfall data are incorporated to assess the influence of precipitation on landslide occurrence. Geological data contribute to the understanding of local geological conditions. Several machine learning algorithms, including but not limited to decision trees, support vector machines, and neural networks, are employed to create predictive models. These models are trained on historical landslide data and validated against real-world cases. Cross-validation techniques are applied to ensure the model’s robustness and generalization capabilities. Artificial Intelligence and Machine Learning Landslide risk Landslide identification Machine learning Deep learning Big data Convolutional neural networks Figures Figure 1 Figure 2 Figure 3 I. INTRODUCTION Landslides represent a significant geohazard that threat- ens communities, infrastructure, and the environment world- wide. These destructive events can be triggered by a variety of factors, including heavy rainfall, geological conditions, and human activities. Timely identification and prediction of landslide-prone areas are essential for disaster mitigation and response efforts[ 1 ]. In recent years, the field of machine learning has emerged as a powerful tool to address this critical challenge, offering the potential to enhance our understanding of landslides and improve our ability to manage their im- pact[ 2 ]. Traditional methods of landslide identification have relied on expert analysis of geological, topographical, and meteo- rological data, often resulting in limited accuracy and the potential for human bias. Machine learning, on the other hand, provides a data-driven approach that can analyze vast datasets, identify complex patterns, and make predictions with a high degree of precision. This has opened up new possibilities for landslide research and risk assessment[ 3 ]. The goal of landslide identification using machine learning is to leverage the vast amount of available data, ranging from satellite imagery and topographic maps to weather records and geological information, to develop predictive models that can identify areas at high risk of landslides. These models can take into account various influencing factors, such as terrain characteristics, land cover, precipitation patterns, and geological composition, to create a comprehensive assessment of landslide susceptibility[ 5 ]. This introduction outlines the importance of landslide iden- tification and highlights the potential of machine learning techniques to revolutionize the field. By harnessing the ca- pabilities of machine learning algorithms, researchers and authorities can gain valuable insights into landslide dynamics, improve preparedness, and implement proactive measures to reduce the impact of these natural disasters[ 6 ]. This paper explores the methodology, data sources, and key machine learning algorithms used in landslide identification, aiming to contribute to more effective landslide risk management and enhanced disaster resilience in vulnerable regions[ 7 ]. The concept of the Internet of Insecure Things (IoT) is slowly becoming relevant across a range of industrial verticals, which raises the value of its economic potential. Process monitoring is changed by IoT through the connection and monitoring of several devices inside a network. Despite all of the benefits of IoT, ensuring network security is quite challeng- ing. Numerous incidents of unauthorised data manipulation and hacking into networked devices, including cameras and autos, highlight the problem[ 8 ]. Nevertheless, the advantages of IoT are outweighed by the enormous problem of guaranteeing network security. The issue is clear given the numerous instances of unauthorised data modification and hacking into networked devices, including cameras and automobiles. However, actual data shows that these tiny IoT devices and sensors are able to run independent instances of tried-and-true, traditional encryption methods. You might be unsure of the definition of encryption. It is the process of encoding information or replacing the original information with a replacement, sometimes referred to as ciphertext. I’ll elaborate on that later[ 9 ]. A landslide is the downward movement of rock, soil, and organic material caused by gravity. A destructive natural calamity, a landslide kills people and severely destroys both plants and property. Due to three primary factors—increased urbanization, deforestation, and precipitation intensity as a result of climate change—it is anticipated to happen more frequently. We are afraid about landslides because of this. The key to lowering the number of fatalities and property losses is landslide prediction and risk assessment. Policymakers, scien- tists, engineers, and the general public can utilize the landslide susceptibility map (LSM), which shows the landslide-prone locations, to prevent catastrophic landslides[ 10 ]. The effectiveness of Machine Learning (ML) in landslide susceptibility mapping is further improved by the availability of readily accessible satellite images, remote sensing data, historical landslide records, and geographical information sys- tems (GIS). This article provides an extensive overview of various ML algorithms used for creating landslide susceptibil- ity maps (LSM), encompassing both traditional and emerging techniques such as conventional, hybrid, ensemble, and deep learning approaches. This survey aims to assist emerging researchers in the selection of ML algorithms, causal factors, performance assessment techniques, and landslide inventories, among other critical considerations[ 11 ]. The study’s contribution encompasses comprehensive re- views that exclusively focus on the utilization of Machine Learning (ML) in landslide susceptibility mapping, aiming to elucidate the intricacies, comparisons, challenges, and future prospects in this field. In their research, Naemitabar et al. con- ducted a comparative analysis of popular ML methodologies employed for LSM generation. The primary emphasis of their study lies in the prioritization of effective landslide causative factors (LCF) to enhance accuracy[ 12 ]. The research reviewed four ML models, including Support Vector Machines (SVM), Boosted Regression Trees (BRT), Logistic Model Tree (LMT), and Random Forest (RF). The findings indicated that SVM, with an Area Under the Receiver Operating Characteristic Curve (AUC) value of 0.86, and RF, with an AUC value of 0.89, exhibited superior performance compared to the others. Several LCFs were investigated, with lithology, slope, slope aspect, distance to fault, and land use/land cover (LULC) emerging as the most effective ones. The study’s author also proposed that prioritizing these effective LCFs for training ML models resulted in increased accuracy[ 13 ]. In a separate study, Zhang et al. employed four traditional ML models, namely Best-First Decision Tree (BFTree), Func- tional Tree (FT), SVM, and Classification Regression Tree (CART). They utilized a bagging-based ensemble approach to enhance the performance of these ML models. This com- parative analysis was conducted in Jiange County, Sichuan Province, China, for the purpose of generating LSM for resource planning and landslide management. The results re- vealed that the bagging-based ensemble method outperformed the conventional ML models. Notably, CART exhibited an AUC value of 0.766 without bagging, which improved to an AUC value of 0.874 with the bagging-CART model[ 14 ]. II. LITERATURE REVIEW Landslides pose a significant threat to both human lives and infrastructure in various parts of the world. Rapid and accurate identification of potential landslide-prone areas is crucial for disaster mitigation and preparedness. In recent years, the integration of machine learning (ML) techniques into landslide identification has gained considerable attention due to its potential to enhance the efficiency and accuracy of landslide assessment and mapping. This literature review provides an overview of recent research and developments in the field of landslide identification using machine learning[15]. A. Machine Learning Approaches for Landslide Identification Machine learning, a subfield of artificial intelligence, en- compasses various algorithms and models that can be em- ployed to identify and predict landslides. Support Vector Machines (SVM), Random Forest, Neural Networks, Decision Trees, and other supervised learning techniques have been extensively explored for their suitability in landslide identi- fication. These algorithms process a range of geospatial data, including remote sensing imagery, topographical information, historical records, and geological factors, to create predictive models that highlight potential landslide-prone areas[16]. B. Feature Selection and Data Integration One of the critical challenges in landslide identification is the selection and integration of relevant features. Re- searchers have developed innovative methods to extract and combine different geospatial and meteorological variables, such as rainfall data, land cover information, soil properties, and slope characteristics. Feature selection techniques, like Principal Component Analysis (PCA) and Recursive Feature Elimination (RFE), aid in improving the accuracy of landslide prediction models[17]. C. Validation and Model Performance Ensuring the reliability and accuracy of ML-based landslide identification models is essential. Researchers have employed various validation techniques, such as k-fold cross-validation and receiver operating characteristic (ROC) analysis, to assess the performance of their models. Metrics like the Area Under the Curve (AUC) and accuracy are often used to quantify a model’s ability to correctly classify landslide-prone areas[18]. D. Landslide Causative Factors and Their Prioritization Understanding the causative factors of landslides is pivotal for accurate identification. Different studies have emphasized the importance of specific variables, including slope, soil type, rainfall intensity, and vegetation cover, in landslide occurrence. Prioritizing these causative factors and incorporating them into the ML models has demonstrated improved accuracy in landslide identification[19]. E. Remote Sensing and Data Acquisition The availability of satellite imagery and remote sensing data has greatly facilitated landslide identification. These sources provide high-resolution and up-to-date information that can be analyzed through ML algorithms. The integration of various satellite sensors, such as optical and synthetic aperture radar (SAR), has offered new insights into landslide monitoring and prediction[20]. III. RESEARCH METHODOLOGY We utilized the Reporting Standards for Systematic Evi- dence Synthesis in Environmental Research (ROSES) for con- ducting our systematic literature survey. ROSES was specif- ically designed to enhance the quality of systematic reviews within the realm of environmental research. It is an extension of PRISMA, a widely recognized reporting standard in the healthcare field. ROSES plays a vital role in ensuring a transparent and reproducible process for conducting literature reviews. In our quest to identify pertinent articles for this survey, we employed various combinations of keywords on the Web of Science (WoS) platform. These keyword combinations in- cluded:“landslide, susceptibility mapping, machine learning”, “landslide, susceptibility mapping, deep learning”, “landslide, susceptibility mapping, SVM, support vector machine”, “land- slide, susceptibility mapping, random forest, RF”, “landslide, susceptibility mapping, hybrid, machine learning”, “landslide, susceptibility mapping, ensemble, machine learning”, “land- slide, susceptibility mapping, ANN, artificial neural network”, and “landslide, susceptibility mapping, DT, decision tree”. We endeavored to encompass a comprehensive array of pop- ular machine learning (ML) models commonly employed in landslide susceptibility mapping, to the best of our knowledge. The initial search, incorporating keywords related to landslide susceptibility mapping and ML, returned a total of 424 papers. Further combinations of keywords, such as coupling ’landslide susceptibility mapping’ with ’SVM’ (231 results), ’RF’ (209 results), ’deep learning’ (67 results), ’hybrid’ (97 results), ’ensemble’ (153 results), ’ANN’ (148 results), and ’DT’ (27 results), produced the specified search result values. In total, our search generated 1356 articles. Subsequently, we refined our selection process using the ROSES reporting standard to filter and identify relevant articles for our study. We implemented specific criteria to eliminate literature that was not pertinent to our study. These criteria were established to ensure the replicability of our results, which is essential for the verification and validation of our survey. In the initial screening process, we scrutinized the titles of articles to deter- mine if they contained the keywords specified in our search. Following this title screening, we identified 143 relevant pa- pers. Subsequently, we removed duplicate titles, reducing the number of articles to 121. The next stage involved retrieving the full text for these 121 filtered articles. Unfortunately, we were unable to obtain one article, and another was not found, resulting in a total of 119 articles available for our study. In the subsequent steps, which encompassed full-text screening and critical appraisal, all 119 articles were found to be pertinent to the subject of landslide susceptibility mapping using machine learning. Therefore, no articles were excluded during these stages of the filtration process. Ultimately, after completing the filtration process, we selected a total of 119 articles for our study. IV. DATA COLLECTION The process of data collection plays a vital role in gathering high-quality evidence, which, in turn, facilitates robust and dependable problem-solving. In our study, confronted with the absence of an available open-source image dataset for landslides, we undertook the task of creating our image dataset. Specifically, we established a binary class dataset, distinguishing between landslide and non-landslide images, as the foundation for addressing the problem at hand. To construct this dataset, we curated a selection of high-quality images (refer to Fig. 2 ) from diverse sources, including Google Images and various websites[ 21 ]. We amassed a robust dataset comprising approximately 260 images, meticulously curated for training and testing our pro- posed deep learning model. In the realm of machine learning, models require training data to learn relevant features, and test data for evaluation. As such, we partitioned our dataset into training and testing subsets. In this partition, 80 percent of the dataset was allocated to the training set within each category, with the remaining 20 percent designated for the test set[ 22 ]. V. MACHINE LEARNING IN LANDSLIDE SUSCEPTIBILITY MAPPING Landslides are geological events that can result in devastat- ing consequences, including loss of life and property damage. Landslide susceptibility mapping is a critical component of disaster risk reduction, as it helps identify areas prone to landslides, enabling proactive measures and informed land use planning. Machine learning techniques have emerged as powerful tools in this field, offering the potential to improve the accuracy of landslide susceptibility mapping. Here are some key aspects of how machine learning is applied in landslide susceptibility mapping: Data Integration : Machine learning models in landslide susceptibility mapping lever- age various data sources. These include topographical data, geological information, land cover data, historical landslide records, rainfall data, and remotely sensed imagery. Integrating these diverse datasets provides a holistic view of the factors influencing landslides[ 23 ]. Feature Engineering One of the critical steps in building machine learning models for landslide susceptibility mapping is feature engineering. Features, representing various geo- logical, topographic, and meteorological characteristics, are extracted from the input data. These features help the models understand the spatial relationships and patterns related to landslide occurrences[ 24 ]. Algorithm Selection Different machine learning algo- rithms are employed for landslide susceptibility mapping, such as decision trees, random forests, support vector machines, and neural networks. The choice of algorithm depends on the complexity of the problem and the nature of the data. Model Training Historical landslide data are used to train the machine learning models. During training, the models learn the relationships between input features and landslide occur- rences, allowing them to make predictions about landslide susceptibility. Validation and Testing : To assess the model’s perfor- mance, it undergoes validation and testing processes. Data that the model has not seen during training are used to evaluate its predictive accuracy. Cross-validation techniques are often employed to ensure the model’s robustness. Spatial Mapping: Once trained and validated, machine learning models generate susceptibility maps that depict areas at different levels of risk. These maps help identify regions highly susceptible to landslides, enabling authorities to prioritize mitigation efforts. Temporal Considerations : Machine learning models can also incorporate temporal aspects, such as changes in rainfall patterns or land cover over time. This dynamic modeling allows for real-time monitoring and adaptation to evolving landslide risks. Uncertainty Assessment: Assessing the un- certainty of susceptibility maps is crucial. Machine learning models can provide probabilistic outputs, indicating the degree of confidence in their predictions. This uncertainty information aids decision-makers in risk management. Integration with Geographic Information Systems (GIS) Machine learning-based landslide susceptibility maps are typically integrated into GIS platforms, making them ac- cessible to planners, policymakers, and emergency responders. GIS tools facilitate spatial analysis and decision-making. Future Enhancements Ongoing research in this field focuses on improving the accuracy and interpretability of machine learning models, as well as their ability to handle data from data-scarce regions and extreme weather events. In conclusion, machine learning plays a pivotal role in enhancing landslide susceptibility mapping by leveraging diverse data sources, advanced algorithms, and spatial modeling. These techniques contribute to better-informed decision-making, im- proved disaster preparedness, and ultimately, the reduction of landslide-related risks in vulnerable regions. As technology advances and more data become available, machine learning will continue to evolve and improve our understanding of landslide dynamics[ 25 ]. VI. RESULTS We conducted a systematic review of the literature and employed specific criteria for article selection, resulting in a final set of 119 relevant articles for our analysis. The selected studies encompassed a range of ML models and methodologies applied to landslide susceptibility mapping. Prevalent Machine Learning Models Our analysis revealed a variety of ML models utilized in landslide identification. Support Vector Machines (SVM), Random Forest (RF), Deep Learning, Hybrid Models, Ensembles, Artificial Neural Net- works (ANN), and Decision Trees (DT) emerged as the most prevalent ML techniques in the surveyed literature. SVM and RF, in particular, demonstrated a noteworthy presence in the reviewed studies, with 231 and 209 instances, respectively. Key Findings from the Literature he literature consistently emphasized the significance of various causative factors in landslide identification. Factors such as slope, soil type, rain- fall intensity, vegetation cover, and distance to fault lines were frequently mentioned as influential in landslide occurrence. Several studies evaluated the performance of ML models using metrics like the Area Under the Curve (AUC) and accuracy. SVM and RF were frequently reported to exhibit better performance compared to other models, with AUC values of 0.86 and 0.89, respectively. The integration of diverse geospatial and meteorological data, including satellite imagery and remote sensing data, played a pivotal role in improv- ing model accuracy. Researchers employed feature selection techniques like Principal Component Analysis (PCA) and Recursive Feature Elimination (RFE) to enhance the relevance of input data. The dataset, which comprises high-resolution photographs of landslides, proved to be invaluable for creating detailed feature maps. A total of 213 images were allocated for training, with the remaining images designated for the test set. Importantly, there was no overlap between these two sets. Our experimental algorithms yielded notable classification accuracies, including Logistic Regression (51.40 percent), Random Forest (80.70 percent), AdaBoost (83.40 percent), K- Nearest Neighbors (KNN) (86.50 percent), and Support Vector Machine (SVM) (92.70 percent). It is worth highlighting that our proposed framework achieved the highest classification accuracy at 96.90 percent, demonstrating an exceptional sen- sitivity and precision rate of 96.90 percent as well. We utilized the ROC curve [ 17 ] to evaluate the performance of our model (Fig. 2 ) in comparison to other machine learning algorithms. Our Proposed Approach demonstrated superior performance with an AUC of 99.20 percent, out- performing SVM (AUC = 92.20 percent), K-Nearest Neighbors (KNN) (AUC = 93.50 percent), AdaBoost (AUC = 83.30 per- cent), Random Forest (RF) (AUC = 86.90 percent), and Logistic Regression (AUC = 50 percent). The deep learning Convolutional Neural Network (CNN) model we proposed exhibited the highest predictive perfor- mance, followed closely by SVM. This can be attributed to the unique design of CNN, which is optimized for processing structured and ordered data. The convolution and max-pooling layers enable weight sharing translationally, resembling the mechanisms of the human visual cortex. This architectural choice has proven to be highly effective in the classification of landslides. VII. CONCLUSION This paper presents an integrated approach for landslide identification through the application of machine learning and deep learning. The methodology is exemplified and validated through a case study conducted on Lantau Island, Hong Kong, using multiple landslide databases. From our investigation, the following conclusions can be drawn: Among the eight machine learning and deep learning models tested (LR, SVM, RF, Discrete AdaBoost, LogitBoost, Gentle Adaboost, CNN-6, and DCNN-11), DCNN-11 emerges as the most promising model for addressing landslide identification challenges. Regarding the three landslide databases, models trained on RecLD exhibit the highest average identification accuracy, achieving 89.3 percent when combined with the DCNN-11 model. On the other hand, the highest accuracy for JLD and ReclLD-trained models stands at 87.5percent and 86.4 percent, respectively. In conclusion, machine learning has revolutionized the field of landslide identification by harnessing the power of data and advanced algorithms. This technology has the potential to save lives, protect communities, and reduce the economic impact of landslides. As we continue to advance our understanding and application of machine learning in landslide identification, we move closer to a future where the devastating consequences of landslides are minimized through proactive risk management and preparedness efforts. References Turner, A.K.; Schuster, R.L. Landslides: Investigation and Mitigation; Special Report 247; National Academy Press: Washington, DC, USA, 1996. [Google Scholar] Fleuchaus, P.; Blum, P.; Wilde, M.; Terhorst, B.; Butscher, C. Retrospec- tive evaluation of landslide susceptibility maps and review of validation practice. Environ. Earth Sci. 2021, 80, 485. [Google Scholar] [CrossRef] Pardeshi, S.D.; Autade, S.E.; Pardeshi, S.S. Landslide hazard assess- ment: Recent trends and techniques. SpringerPlus 2013, 2, 523. [Google Scholar] [CrossRef] [PubMed][Green Version] Huabin, W.; Gangjun, L.; Weiya, X.; Gonghui, W. GIS-based landslide hazard assessment: An overview. Prog. Phys. Geogr. 2005, 29, 548–567. [Google Scholar] [CrossRef] Reichenbach, P.; Rossi, M.; Malamud, B.D.; Mihir, M.; Guzzetti, F. A review of statistically-based landslide susceptibility models. Earth-Sci. Rev. 2018, 180, 60–91. [Google Scholar] [CrossRef] Arabameri, A.; Rezaei, K.; Pourghasemi, H.R.; Lee, S.; Yamani, M. GIS-based gully erosion susceptibility mapping: A comparison among three data-driven models and AHP knowledge-based technique. Environ. Earth Sci. 2018, 77, 628. [Google Scholar] [CrossRef] Merghadi, A.; Yunus, A.P.; Dou, J.; Whiteley, J.; ThaiPham, B.; Bui, D.T.; Avtar, R.; Abderrahmane, B. Machine learning methods for landslide susceptibility studies: A comparative overview of algorithm performance. Earth-Sci. Rev. 2020, 207, 103225. [Google Scholar] [CrossRef] Huang, Y.; Zhao, L. Review on landslide susceptibility mapping using support vector machines. Catena 2018, 165, 520–529. [Google Scholar] [CrossRef] Naemitabar, M.; Zanganeh Asadi, M. Landslide zonation and assessment of Farizi watershed in northeastern Iran using data mining techniques. Nat. Hazards 2021, 108, 2423–2453. [Google Scholar] [CrossRef] Zhang, T.; Fu, Q.; Wang, H.; Liu, F.; Wang, H.; Han, L. Bagging-based machine learning algorithms for landslide susceptibility modeling. Nat. Hazards 2021, 110, 823–846. [Google Scholar] [CrossRef] Clarivate. Web of Science: Discover Multidisciplinary Content. Avail- able online: https://www.webofscience.com/wos/woscc/basic-search (ac- cessed on 2 March 2022). Saha, S.; Saha, A.; Hembram, T.K.; Pradhan, B.; Alamri, A.M. Eval- uating the performance of individual and novel ensemble of machine learning and statistical models for landslide susceptibility assessment at Rudraprayag District of Garhwal Himalaya. Appl. Sci. 2020, 10, 3772. [Google Scholar] [CrossRef] Fang, Z.; Wang, Y.; Peng, L.; Hong, H. Integration of convolutional neu- ral network and conventional machine learning classifiers for landslide susceptibility mapping. Comput. Geosci. 2020, 139, 104470. [Google Scholar] [CrossRef] Pham, B.T.; Bui, D.T.; Prakash, I.; Dholakia, M. Hybrid integration of Multilayer Perceptron Neural Networks and machine learning ensembles for landslide susceptibility assessment at Himalayan area (India) using GIS. Catena 2017, 149, 52–63. [Google Scholar] [CrossRef] Haddaway, N.R.; Macura, B.; Whaley, P.; Pullin, A.S. ROSES RepOrting standards for Systematic Evidence Syntheses: Pro forma, flow-diagram and descriptive summary of the plan and conduct of environmental systematic reviews and systematic maps. Environ. Evid. 2018, 7, 7. [Google Scholar] [CrossRef] Moher, D.; Liberati, A.; Tetzlaff, J.; Altman, D.G.; PRISMA Group. Preferred reporting items for systematic reviews and meta-analyses: The PRISMA statement. Ann. Intern. Med. 2009, 151, 264–269. [Google Scholar] [CrossRef] [PubMed][Green Version] Pradhan, B. A comparative study on the predictive ability of the decision tree, support vector machine and neuro-fuzzy models in landslide susceptibility mapping using GIS. Comput. Geosci. 2013, 51, 350–365. [Google Scholar] [CrossRef] Yilmaz, I. Landslide susceptibility mapping using frequency ratio, logistic regression, artificial neural networks and their comparison: A case study from Kat landslides (Tokat—Turkey). Comput. Geosci. 2009, 35, 1125–1138. [Google Scholar] [CrossRef] Yilmaz, I. Comparison of landslide susceptibility mapping methodolo- gies for Koyulhisar, Turkey: Conditional probability, logistic regression, artificial neural networks, and support vector machine. Environ. Earth Sci. 2010, 61, 821–836. [Google Scholar] [CrossRef] Youssef, A.M.; Pourghasemi, H.R.; Pourtaghi, Z.S.; Al-Katheeri, M.M. Landslide susceptibility mapping using random forest, boosted regres- sion tree, classification and regression tree, and general linear models and comparison of their performance at Wadi Tayyah Basin, Asir Region, Saudi Arabia. Landslides 2016, 13, 839–856. [Google Scholar] [CrossRef] Yao, X.; Tham, L.; Dai, F. Landslide susceptibility mapping based on support vector machine: A case study on natural slopes of Hong Kong, China. Geomorphology 2008, 101, 572–582. [Google Scholar] [CrossRef] Zare, M.; Pourghasemi, H.R.; Vafakhah, M.; Pradhan, B. Landslide susceptibility mapping at Vaz Watershed (Iran) using an artificial neural network model: A comparison between multilayer perceptron (MLP) and radial basic function (RBF) algorithms. Arab. J. Geosci. 2013, 6, 2873–2888. [Google Scholar] [CrossRef] Cruden, D.M., 1991. A simple definition of a landslide. Bulletin of Engineering Geology and the Environment, 43(1), pp.27-29 Parkash, S., 2011. Historical records of socio-economically significant landslides in India. J South Asia Disaster Studies, 4(2), pp.177-204. Jime´nez-Pera´lvarez, J.D., El Hamdouni, R., Palenzuela, J.A., Irigaray, C. and Chaco´n, J., 2017. Landslide-hazard mapping through multi- technique activity assessment: an example from the Betic Cordillera (southern Spain). Landslides, 14(6), pp.1975-1991 Additional Declarations The authors declare no competing interests. Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-4632694","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":318550584,"identity":"edb05b50-3613-4433-9c20-971c8947e191","order_by":0,"name":"Karan Sarawagi","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA/ElEQVRIiWNgGAWjYBACCQSTueFAQgWIZm4gVgtj44MHZ0BaGInX0mz4sA3MwK9Fsv3sMYkfv+zyDY4fbJNInFcbzd8O1PKjYhtOLdI8eWmSvX3JlhvOJAK1bDueO+MwYwNjz5nbOLXIMeSYSfD2MBuYHQBrOZbbANTCzNiGRwv/GzPJvz31BmbnHwK1zDmWO5+QFmmJHDNpnh+HDcxuJDYbJDbU5G4gpEVyxhtja9mG4wb2Nx42Pkg4diB3I1DLQXx+kTifY3jzzZ9qA8n+5AMHf9TU5c47f/jggx8VuLWAAWMbnHkYTB7Arx4E/sBZdYQVj4JRMApGwYgDAFTIYZHxw/7rAAAAAElFTkSuQmCC","orcid":"https://orcid.org/0009-0007-4800-275X","institution":"Chandigarh University","correspondingAuthor":true,"prefix":"","firstName":"Karan","middleName":"","lastName":"Sarawagi","suffix":""},{"id":318550585,"identity":"ecb4aa2d-bbdb-4234-834d-73140877b68d","order_by":1,"name":"Navjot Singh","email":"","orcid":"","institution":"Chandigarh University","correspondingAuthor":false,"prefix":"","firstName":"Navjot","middleName":"","lastName":"Singh","suffix":""},{"id":318550586,"identity":"a28adb07-255e-4899-8724-87a51e5e8c1e","order_by":2,"name":"Khushwant Virdi","email":"","orcid":"","institution":"Chandigarh University","correspondingAuthor":false,"prefix":"","firstName":"Khushwant","middleName":"","lastName":"Virdi","suffix":""}],"badges":[],"createdAt":"2024-06-25 01:15:03","currentVersionCode":1,"declarations":{"humanSubjects":false,"vertebrateSubjects":false,"conflictsOfInterestStatement":false,"humanSubjectEthicalGuidelines":false,"humanSubjectConsent":false,"humanSubjectClinicalTrial":false,"humanSubjectCaseReport":false,"vertebrateSubjectEthicalGuidelines":false},"doi":"10.21203/rs.3.rs-4632694/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-4632694/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":59120367,"identity":"1367aa83-91ab-4876-9aad-bdb6bcf49877","added_by":"auto","created_at":"2024-06-26 14:46:47","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":140847,"visible":true,"origin":"","legend":"\u003cp\u003eOverview of Model\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-4632694/v1/a986c7a55529035197d068bd.png"},{"id":59120369,"identity":"34290f41-ee1a-49d9-b8a0-531401c6a6ea","added_by":"auto","created_at":"2024-06-26 14:46:47","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":849742,"visible":true,"origin":"","legend":"\u003cp\u003eSample landslide images extracted from our dataset\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-4632694/v1/25e16601f6c6a0dcd3aafd2f.png"},{"id":59120368,"identity":"1f4860db-4ae1-4cb6-9c6f-3ab7a83f68b8","added_by":"auto","created_at":"2024-06-26 14:46:47","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":274891,"visible":true,"origin":"","legend":"\u003cp\u003eWe generated an ROC curve illustrating different models, including the proposed approach\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-4632694/v1/9feb5cb6da14bd2db878211d.png"},{"id":59120381,"identity":"a898a069-e910-4dc7-9039-6e96d1c28441","added_by":"auto","created_at":"2024-06-26 14:46:52","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1451888,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-4632694/v1/ec1b670c-2ef3-4d66-bdd5-adcb6c5b180c.pdf"}],"financialInterests":"The authors declare no competing interests.","formattedTitle":"\u003cp\u003ePredicting Landslides with Machine Learning: A Data-Driven Approach\u003c/p\u003e","fulltext":[{"header":"I. INTRODUCTION","content":"\u003cp\u003eLandslides represent a significant geohazard that threat- ens communities, infrastructure, and the environment world- wide. These destructive events can be triggered by a variety of factors, including heavy rainfall, geological conditions, and human activities. Timely identification and prediction of landslide-prone areas are essential for disaster mitigation and response efforts[\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]. In recent years, the field of machine learning has emerged as a powerful tool to address this critical challenge, offering the potential to enhance our understanding of landslides and improve our ability to manage their im- pact[\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eTraditional methods of landslide identification have relied on expert analysis of geological, topographical, and meteo- rological data, often resulting in limited accuracy and the potential for human bias. Machine learning, on the other hand, provides a data-driven approach that can analyze vast datasets, identify complex patterns, and make predictions with a high degree of precision. This has opened up new possibilities for landslide research and risk assessment[\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eThe goal of landslide identification using machine learning is to leverage the vast amount of available data, ranging from satellite imagery and topographic maps to weather records and geological information, to develop predictive models that can identify areas at high risk of landslides. These models can take into account various influencing factors, such as terrain characteristics, land cover, precipitation patterns, and geological composition, to create a comprehensive assessment of landslide susceptibility[\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eThis introduction outlines the importance of landslide iden- tification and highlights the potential of machine learning techniques to revolutionize the field. By harnessing the ca- pabilities of machine learning algorithms, researchers and authorities can gain valuable insights into landslide dynamics, improve preparedness, and implement proactive measures to reduce the impact of these natural disasters[\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e]. This paper explores the methodology, data sources, and key machine learning algorithms used in landslide identification, aiming to contribute to more effective landslide risk management and enhanced disaster resilience in vulnerable regions[\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eThe concept of the Internet of Insecure Things (IoT) is slowly becoming relevant across a range of industrial verticals, which raises the value of its economic potential. Process monitoring is changed by IoT through the connection and monitoring of several devices inside a network. Despite all of the benefits of IoT, ensuring network security is quite challeng- ing. Numerous incidents of unauthorised data manipulation and hacking into networked devices, including cameras and autos, highlight the problem[\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eNevertheless, the advantages of IoT are outweighed by the enormous problem of guaranteeing network security. The issue\u003c/p\u003e \u003cp\u003eis clear given the numerous instances of unauthorised data modification and hacking into networked devices, including cameras and automobiles. However, actual data shows that these tiny IoT devices and sensors are able to run independent instances of tried-and-true, traditional encryption methods. You might be unsure of the definition of encryption. It is the process of encoding information or replacing the original information with a replacement, sometimes referred to as ciphertext. I\u0026rsquo;ll elaborate on that later[\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eA landslide is the downward movement of rock, soil, and organic material caused by gravity. A destructive natural calamity, a landslide kills people and severely destroys both plants and property. Due to three primary factors\u0026mdash;increased urbanization, deforestation, and precipitation intensity as a result of climate change\u0026mdash;it is anticipated to happen more frequently. We are afraid about landslides because of this. The key to lowering the number of fatalities and property losses is landslide prediction and risk assessment. Policymakers, scien- tists, engineers, and the general public can utilize the landslide susceptibility map (LSM), which shows the landslide-prone locations, to prevent catastrophic landslides[\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eThe effectiveness of Machine Learning (ML) in landslide susceptibility mapping is further improved by the availability of readily accessible satellite images, remote sensing data, historical landslide records, and geographical information sys- tems (GIS). This article provides an extensive overview of various ML algorithms used for creating landslide susceptibil- ity maps (LSM), encompassing both traditional and emerging techniques such as conventional, hybrid, ensemble, and deep learning approaches. This survey aims to assist emerging researchers in the selection of ML algorithms, causal factors, performance assessment techniques, and landslide inventories, among other critical considerations[\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eThe study\u0026rsquo;s contribution encompasses comprehensive re- views that exclusively focus on the utilization of Machine Learning (ML) in landslide susceptibility mapping, aiming to elucidate the intricacies, comparisons, challenges, and future prospects in this field. In their research, Naemitabar et al. con- ducted a comparative analysis of popular ML methodologies employed for LSM generation. The primary emphasis of their study lies in the prioritization of effective landslide causative factors (LCF) to enhance accuracy[\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e]. The research reviewed four ML models, including Support Vector Machines (SVM), Boosted Regression Trees (BRT), Logistic Model Tree (LMT), and Random Forest (RF). The findings indicated that SVM, with an Area Under the Receiver Operating Characteristic Curve (AUC) value of 0.86, and RF, with an AUC value of 0.89, exhibited superior performance compared to the others. Several LCFs were investigated, with lithology, slope, slope aspect, distance to fault, and land use/land cover (LULC) emerging as the most effective ones. The study\u0026rsquo;s author also proposed that prioritizing these effective LCFs for training ML models resulted in increased accuracy[\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eIn a separate study, Zhang et al. employed four traditional ML models, namely Best-First Decision Tree (BFTree), Func- tional Tree (FT), SVM, and Classification Regression Tree\u003c/p\u003e \u003cp\u003e(CART). They utilized a bagging-based ensemble approach to enhance the performance of these ML models. This com- parative analysis was conducted in Jiange County, Sichuan Province, China, for the purpose of generating LSM for resource planning and landslide management. The results re- vealed that the bagging-based ensemble method outperformed the conventional ML models. Notably, CART exhibited an AUC value of 0.766 without bagging, which improved to an AUC value of 0.874 with the bagging-CART model[\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e].\u003c/p\u003e"},{"header":"II. LITERATURE REVIEW","content":"\u003cp\u003eLandslides pose a significant threat to both human lives\u0026nbsp;and infrastructure in various parts of the world. Rapid and accurate identification of potential landslide-prone areas is crucial for disaster mitigation and preparedness. In recent years, the integration of machine learning (ML) techniques into landslide identification has gained considerable attention due\u0026nbsp;to\u0026nbsp;its\u0026nbsp;potential\u0026nbsp;to\u0026nbsp;enhance\u0026nbsp;the\u0026nbsp;efficiency\u0026nbsp;and\u0026nbsp;accuracy of landslide assessment and mapping. This literature review provides an overview of recent research and developments in the\u0026nbsp;field\u0026nbsp;of\u0026nbsp;landslide\u0026nbsp;identification\u0026nbsp;using\u0026nbsp;machine\u0026nbsp;learning[15].\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eA.\u0026nbsp;\u0026nbsp;\u003c/em\u003e\u003cem\u003e\u0026nbsp;\u0026nbsp;Machine Learning Approaches for Landslide Identification\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eMachine learning, a subfield of artificial intelligence, en- compasses various algorithms and models that can be em- ployed to identify and predict landslides. Support Vector Machines\u0026nbsp;(SVM),\u0026nbsp;Random\u0026nbsp;Forest,\u0026nbsp;Neural\u0026nbsp;Networks,\u0026nbsp;Decision Trees, and other supervised learning techniques have been extensively explored for their suitability in landslide identi- fication. These algorithms process a range of geospatial data, including remote sensing imagery, topographical information, historical records, and geological factors, to create predictive models that highlight potential landslide-prone areas[16].\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eB.\u0026nbsp; \u0026nbsp;\u003c/em\u003e\u003cem\u003eFeature\u0026nbsp;Selection\u0026nbsp;and\u0026nbsp;Data\u0026nbsp;Integration\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eOne\u0026nbsp;of\u0026nbsp;the\u0026nbsp;critical\u0026nbsp;challenges\u0026nbsp;in\u0026nbsp;landslide\u0026nbsp;identification is the selection and integration of relevant features. Re- searchers have developed innovative methods to extract and combine different geospatial and meteorological variables, such as rainfall data, land cover information, soil properties, and slope characteristics. Feature selection techniques, like Principal Component Analysis (PCA) and Recursive Feature Elimination (RFE), aid in improving the accuracy of landslide prediction models[17].\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eC.\u0026nbsp; \u0026nbsp;\u003c/em\u003e\u003cem\u003eValidation\u0026nbsp;and\u0026nbsp;Model\u0026nbsp;Performance\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eEnsuring\u0026nbsp;the\u0026nbsp;reliability\u0026nbsp;and\u0026nbsp;accuracy\u0026nbsp;of\u0026nbsp;ML-based\u0026nbsp;landslide identification models is essential. Researchers have employed various validation techniques, such as k-fold cross-validation and\u0026nbsp;receiver\u0026nbsp;operating\u0026nbsp;characteristic\u0026nbsp;(ROC)\u0026nbsp;analysis,\u0026nbsp;to\u0026nbsp;assess the performance of their models. Metrics like the Area Under the Curve (AUC) and accuracy are often used to quantify a model’s\u0026nbsp;ability\u0026nbsp;to\u0026nbsp;correctly\u0026nbsp;classify\u0026nbsp;landslide-prone\u0026nbsp;areas[18].\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eD.\u0026nbsp; \u0026nbsp;\u003c/em\u003e\u003cem\u003eLandslide\u0026nbsp;Causative\u0026nbsp;Factors\u0026nbsp;and\u0026nbsp;Their\u0026nbsp;Prioritization\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eUnderstanding the causative factors of landslides is pivotal for accurate identification. Different studies have emphasized the\u0026nbsp;importance\u0026nbsp;of\u0026nbsp;specific\u0026nbsp;variables,\u0026nbsp;including\u0026nbsp;slope,\u0026nbsp;soil\u0026nbsp;type, rainfall\u0026nbsp;intensity,\u0026nbsp;and\u0026nbsp;vegetation\u0026nbsp;cover,\u0026nbsp;in\u0026nbsp;landslide\u0026nbsp;occurrence. Prioritizing these causative factors and incorporating them\u0026nbsp;into the ML models has demonstrated improved accuracy in landslide identification[19].\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eE.\u0026nbsp; \u0026nbsp;\u003c/em\u003e\u003cem\u003eRemote\u0026nbsp;Sensing\u0026nbsp;and\u0026nbsp;Data\u0026nbsp;Acquisition\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eThe availability of satellite imagery and remote sensing data has greatly facilitated landslide identification. These sources provide high-resolution and up-to-date information that can be analyzed through ML algorithms. The integration of various satellite sensors, such as optical and synthetic aperture radar (SAR), has offered new insights into landslide monitoring and prediction[20].\u003c/p\u003e"},{"header":"III. RESEARCH METHODOLOGY","content":"\u003cp\u003eWe utilized the Reporting Standards for Systematic Evi- dence Synthesis in Environmental Research (ROSES) for con- ducting our systematic literature survey. ROSES was specif- ically designed to enhance the quality of systematic reviews within the realm of environmental research. It is an extension of PRISMA, a widely recognized reporting standard in the healthcare field. ROSES plays a vital role in ensuring a transparent and reproducible process for conducting literature reviews.\u003c/p\u003e \u003cp\u003eIn our quest to identify pertinent articles for this survey, we employed various combinations of keywords on the Web of Science (WoS) platform. These keyword combinations in- cluded:\u0026ldquo;landslide, susceptibility mapping, machine learning\u0026rdquo;, \u0026ldquo;landslide, susceptibility mapping, deep learning\u0026rdquo;, \u0026ldquo;landslide, susceptibility mapping, SVM, support vector machine\u0026rdquo;, \u0026ldquo;land- slide, susceptibility mapping, random forest, RF\u0026rdquo;, \u0026ldquo;landslide, susceptibility mapping, hybrid, machine learning\u0026rdquo;, \u0026ldquo;landslide, susceptibility mapping, ensemble, machine learning\u0026rdquo;, \u0026ldquo;land- slide, susceptibility mapping, ANN, artificial neural network\u0026rdquo;, and \u0026ldquo;landslide, susceptibility mapping, DT, decision tree\u0026rdquo;.\u003c/p\u003e \u003cp\u003eWe endeavored to encompass a comprehensive array of pop- ular machine learning (ML) models commonly employed in landslide susceptibility mapping, to the best of our knowledge. The initial search, incorporating keywords related to landslide susceptibility mapping and ML, returned a total of 424 papers. Further combinations of keywords, such as coupling \u0026rsquo;landslide susceptibility mapping\u0026rsquo; with \u0026rsquo;SVM\u0026rsquo; (231 results), \u0026rsquo;RF\u0026rsquo; (209 results), \u0026rsquo;deep learning\u0026rsquo; (67 results), \u0026rsquo;hybrid\u0026rsquo; (97 results), \u0026rsquo;ensemble\u0026rsquo; (153 results), \u0026rsquo;ANN\u0026rsquo; (148 results), and \u0026rsquo;DT\u0026rsquo; (27 results), produced the specified search result values. In total, our search generated 1356 articles. Subsequently, we refined our selection process using the ROSES reporting standard to filter and identify relevant articles for our study.\u003c/p\u003e \u003cp\u003eWe implemented specific criteria to eliminate literature that was not pertinent to our study. These criteria were established to ensure the replicability of our results, which is essential for the verification and validation of our survey. In the initial\u003c/p\u003e \u003cp\u003escreening process, we scrutinized the titles of articles to deter- mine if they contained the keywords specified in our search. Following this title screening, we identified 143 relevant pa- pers. Subsequently, we removed duplicate titles, reducing the number of articles to 121. The next stage involved retrieving the full text for these 121 filtered articles. Unfortunately, we were unable to obtain one article, and another was not found, resulting in a total of 119 articles available for our study.\u003c/p\u003e \u003cp\u003eIn the subsequent steps, which encompassed full-text screening and critical appraisal, all 119 articles were found to be pertinent to the subject of landslide susceptibility mapping using machine learning. Therefore, no articles were excluded during these stages of the filtration process. Ultimately, after completing the filtration process, we selected a total of 119 articles for our study.\u003c/p\u003e"},{"header":"IV. DATA COLLECTION","content":"\u003cp\u003eThe process of data collection plays a vital role in gathering high-quality evidence, which, in turn, facilitates robust and dependable problem-solving. In our study, confronted with the absence of an available open-source image dataset for landslides, we undertook the task of creating our image dataset. Specifically, we established a binary class dataset, distinguishing between landslide and non-landslide images, as the foundation for addressing the problem at hand. To construct this dataset, we curated a selection of high-quality images (refer to Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e) from diverse sources, including Google Images and various websites[\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eWe amassed a robust dataset comprising approximately 260 images, meticulously curated for training and testing our pro- posed deep learning model. In the realm of machine learning, models require training data to learn relevant features, and test data for evaluation. As such, we partitioned our dataset into training and testing subsets. In this partition, 80 percent of the dataset was allocated to the training set within each category, with the remaining 20 percent designated for the test set[\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e].\u003c/p\u003e "},{"header":"V. MACHINE LEARNING IN LANDSLIDE SUSCEPTIBILITY MAPPING","content":"\u003cp\u003eLandslides are geological events that can result in devastat- ing consequences, including loss of life and property damage. Landslide susceptibility mapping is a critical component of disaster risk reduction, as it helps identify areas prone to landslides, enabling proactive measures and informed land use planning. Machine learning techniques have emerged as powerful tools in this field, offering the potential to improve the accuracy of landslide susceptibility mapping. Here are some key aspects of how machine learning is applied in landslide susceptibility mapping: \u003cstrong\u003eData Integration\u003c/strong\u003e: Machine learning models in landslide susceptibility mapping lever- age various data sources. These include topographical data, geological information, land cover data, historical landslide records, rainfall data, and remotely sensed imagery. Integrating these diverse datasets provides a holistic view of the factors influencing landslides[\u003cspan class=\"CitationRef\"\u003e23\u003c/span\u003e].\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFeature Engineering\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eOne of the critical steps in building machine learning models for landslide susceptibility mapping is feature engineering. Features, representing various geo- logical, topographic, and meteorological characteristics, are extracted from the input data. These features help the models understand the spatial relationships and patterns related to landslide occurrences[\u003cspan class=\"CitationRef\"\u003e24\u003c/span\u003e].\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAlgorithm Selection\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eDifferent machine learning algo- rithms are employed for landslide susceptibility mapping, such as decision trees, random forests, support vector machines, and neural networks. The choice of algorithm depends on the complexity of the problem and the nature of the data.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eModel Training\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eHistorical landslide data are used to train the machine learning models. During training, the models learn the relationships between input features and landslide occur- rences, allowing them to make predictions about landslide susceptibility.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eValidation and Testing\u003c/strong\u003e: To assess the model\u0026rsquo;s perfor- mance, it undergoes validation and testing processes. Data that the model has not seen during training are used to evaluate its predictive accuracy. Cross-validation techniques are often employed to ensure the model\u0026rsquo;s robustness. Spatial Mapping: Once trained and validated, machine learning models generate susceptibility maps that depict areas at different levels of risk. These maps help identify regions highly susceptible to landslides, enabling authorities to prioritize mitigation efforts. \u003cstrong\u003eTemporal Considerations\u003c/strong\u003e: Machine learning models can also incorporate temporal aspects, such as changes in rainfall patterns or land cover over time. This dynamic modeling allows for real-time monitoring and adaptation to evolving landslide risks. Uncertainty Assessment: Assessing the un- certainty of susceptibility maps is crucial. Machine learning models can provide probabilistic outputs, indicating the degree of confidence in their predictions. This uncertainty information aids decision-makers in risk management.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eIntegration with Geographic Information Systems (GIS)\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eMachine learning-based landslide susceptibility maps are typically integrated into GIS platforms, making them ac- cessible to planners, policymakers, and emergency responders. GIS tools facilitate spatial analysis and decision-making.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFuture Enhancements\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eOngoing research in this field focuses on improving the accuracy and interpretability of machine learning models, as well as their ability to handle data from data-scarce regions and extreme weather events. In conclusion, machine learning plays a pivotal role in enhancing landslide susceptibility mapping by leveraging diverse data sources, advanced algorithms, and spatial modeling. These techniques contribute to better-informed decision-making, im- proved disaster preparedness, and ultimately, the reduction of landslide-related risks in vulnerable regions. As technology advances and more data become available, machine learning will continue to evolve and improve our understanding of landslide dynamics[\u003cspan class=\"CitationRef\"\u003e25\u003c/span\u003e].\u003c/p\u003e"},{"header":"VI. RESULTS","content":"\u003cp\u003eWe conducted a systematic review of the literature and employed specific criteria for article selection, resulting in a final set of 119 relevant articles for our analysis. The selected studies encompassed a range of ML models and methodologies applied to landslide susceptibility mapping. \u003cstrong\u003ePrevalent Machine Learning Models\u003c/strong\u003e Our analysis revealed a variety of ML models utilized in landslide identification. Support Vector Machines (SVM), Random Forest (RF), Deep Learning, Hybrid Models, Ensembles, Artificial Neural Net- works (ANN), and Decision Trees (DT) emerged as the most prevalent ML techniques in the surveyed literature. SVM and RF, in particular, demonstrated a noteworthy presence in the reviewed studies, with 231 and 209 instances, respectively. \u003cstrong\u003eKey Findings from the Literature\u003c/strong\u003e he literature consistently emphasized the significance of various causative factors in landslide identification. Factors such as slope, soil type, rain- fall intensity, vegetation cover, and distance to fault lines were frequently mentioned as influential in landslide occurrence. Several studies evaluated the performance of ML models using metrics like the Area Under the Curve (AUC) and accuracy. SVM and RF were frequently reported to exhibit better performance compared to other models, with AUC values of 0.86 and 0.89, respectively. The integration of diverse geospatial and meteorological data, including satellite imagery and remote sensing data, played a pivotal role in improv- ing model accuracy. Researchers employed feature selection techniques like Principal Component Analysis (PCA) and Recursive Feature Elimination (RFE) to enhance the relevance of input data.\u003c/p\u003e\n\u003cp\u003eThe dataset, which comprises high-resolution photographs of landslides, proved to be invaluable for creating detailed feature maps. A total of 213 images were allocated for training, with the remaining images designated for the test set. Importantly, there was no overlap between these two sets.\u003c/p\u003e\n\u003cp\u003eOur experimental algorithms yielded notable classification accuracies, including Logistic Regression (51.40 percent), Random Forest (80.70 percent), AdaBoost (83.40 percent), K- Nearest Neighbors (KNN) (86.50 percent), and Support Vector Machine (SVM) (92.70 percent). It is worth highlighting that our proposed framework achieved the highest classification accuracy at 96.90 percent, demonstrating an exceptional sen- sitivity and precision rate of 96.90 percent as well.\u003c/p\u003e\n\u003cp\u003eWe utilized the ROC curve [\u003cspan class=\"CitationRef\"\u003e17\u003c/span\u003e] to evaluate the performance of our model (Fig. \u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003e) in comparison to other machine learning algorithms. Our Proposed Approach demonstrated superior performance with an AUC of 99.20 percent, out- performing SVM (AUC\u0026thinsp;=\u0026thinsp;92.20 percent), K-Nearest Neighbors (KNN) (AUC\u0026thinsp;=\u0026thinsp;93.50 percent), AdaBoost (AUC\u0026thinsp;=\u0026thinsp;83.30 per- cent), Random Forest (RF) (AUC\u0026thinsp;=\u0026thinsp;86.90 percent), and Logistic Regression (AUC\u0026thinsp;=\u0026thinsp;50 percent).\u003c/p\u003e\n\u003cp\u003eThe deep learning Convolutional Neural Network (CNN) model we proposed exhibited the highest predictive perfor- mance, followed closely by SVM. This can be attributed to the unique design of CNN, which is optimized for processing structured and ordered data. The convolution and max-pooling layers enable weight sharing translationally, resembling the mechanisms of the human visual cortex. This architectural choice has proven to be highly effective in the classification of landslides.\u003c/p\u003e"},{"header":"VII. CONCLUSION","content":"\u003cp\u003eThis paper presents an integrated approach for landslide identification through the application of machine learning and deep learning. The methodology is exemplified and validated through a case study conducted on Lantau Island, Hong Kong, using multiple landslide databases. From our investigation, the following conclusions can be drawn: Among the eight machine learning and deep learning models tested (LR, SVM, RF, Discrete AdaBoost, LogitBoost, Gentle Adaboost, CNN-6, and DCNN-11), DCNN-11 emerges as the most promising model for addressing landslide identification challenges. Regarding the three landslide databases, models trained on RecLD exhibit the highest average identification accuracy, achieving 89.3 percent when combined with the DCNN-11 model. On the other hand, the highest accuracy for JLD and ReclLD-trained models stands at 87.5percent and 86.4 percent, respectively.\u003c/p\u003e \u003cp\u003eIn conclusion, machine learning has revolutionized the field of landslide identification by harnessing the power of data and advanced algorithms. This technology has the potential to save lives, protect communities, and reduce the economic impact of landslides. As we continue to advance our understanding and application of machine learning in landslide identification, we move closer to a future where the devastating consequences of landslides are minimized through proactive risk management and preparedness efforts.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eTurner, A.K.; Schuster, R.L. Landslides: Investigation and Mitigation; Special Report 247; National Academy Press: Washington, DC, USA, 1996. [Google Scholar]\u003c/li\u003e\n\u003cli\u003eFleuchaus, P.; Blum, P.; Wilde, M.; Terhorst, B.; Butscher, C. Retrospec- tive evaluation of landslide susceptibility maps and review of validation practice. Environ. Earth Sci. 2021, 80, 485. [Google Scholar] [CrossRef]\u003c/li\u003e\n\u003cli\u003ePardeshi, S.D.; Autade, S.E.; Pardeshi, S.S. Landslide hazard assess- ment: Recent trends and techniques. SpringerPlus 2013, 2, 523. [Google Scholar] [CrossRef] [PubMed][Green Version]\u003c/li\u003e\n\u003cli\u003eHuabin, W.; Gangjun, L.; Weiya, X.; Gonghui, W. GIS-based landslide hazard assessment: An overview. Prog. Phys. Geogr. 2005, 29, 548\u0026ndash;567. [Google Scholar] [CrossRef]\u003c/li\u003e\n\u003cli\u003eReichenbach, P.; Rossi, M.; Malamud, B.D.; Mihir, M.; Guzzetti, F. A review of statistically-based landslide susceptibility models. Earth-Sci. Rev. 2018, 180, 60\u0026ndash;91. [Google Scholar] [CrossRef]\u003c/li\u003e\n\u003cli\u003eArabameri, A.; Rezaei, K.; Pourghasemi, H.R.; Lee, S.; Yamani, M. GIS-based gully erosion susceptibility mapping: A comparison among three data-driven models and AHP knowledge-based technique. Environ. Earth Sci. 2018, 77, 628. [Google Scholar] [CrossRef]\u003c/li\u003e\n\u003cli\u003eMerghadi, A.; Yunus, A.P.; Dou, J.; Whiteley, J.; ThaiPham, B.; Bui, D.T.; Avtar, R.; Abderrahmane, B. Machine learning methods for landslide susceptibility studies: A comparative overview of algorithm performance. Earth-Sci. Rev. 2020, 207, 103225. [Google Scholar] [CrossRef]\u003c/li\u003e\n\u003cli\u003eHuang, Y.; Zhao, L. Review on landslide susceptibility mapping using support vector machines. Catena 2018, 165, 520\u0026ndash;529. [Google Scholar] [CrossRef]\u003c/li\u003e\n\u003cli\u003eNaemitabar, M.; Zanganeh Asadi, M. Landslide zonation and assessment of Farizi watershed in northeastern Iran using data mining techniques. Nat. Hazards 2021, 108, 2423\u0026ndash;2453. [Google Scholar] [CrossRef]\u003c/li\u003e\n\u003cli\u003eZhang, T.; Fu, Q.; Wang, H.; Liu, F.; Wang, H.; Han, L. Bagging-based machine learning algorithms for landslide susceptibility modeling. Nat. Hazards 2021, 110, 823\u0026ndash;846. [Google Scholar] [CrossRef]\u003c/li\u003e\n\u003cli\u003eClarivate. Web of Science: Discover Multidisciplinary Content. Avail- able online: https://www.webofscience.com/wos/woscc/basic-search (ac- cessed on 2 March 2022).\u003c/li\u003e\n\u003cli\u003eSaha, S.; Saha, A.; Hembram, T.K.; Pradhan, B.; Alamri, A.M. Eval- uating the performance of individual and novel ensemble of machine learning and statistical models for landslide susceptibility assessment at Rudraprayag District of Garhwal Himalaya. Appl. Sci. 2020, 10, 3772. [Google Scholar] [CrossRef]\u003c/li\u003e\n\u003cli\u003eFang, Z.; Wang, Y.; Peng, L.; Hong, H. Integration of convolutional neu- ral network and conventional machine learning classifiers for landslide susceptibility mapping. Comput. Geosci. 2020, 139, 104470. [Google Scholar] [CrossRef]\u003c/li\u003e\n\u003cli\u003ePham, B.T.; Bui, D.T.; Prakash, I.; Dholakia, M. Hybrid integration of Multilayer Perceptron Neural Networks and machine learning ensembles for landslide susceptibility assessment at Himalayan area (India) using GIS. Catena 2017, 149, 52\u0026ndash;63. [Google Scholar] [CrossRef]\u003c/li\u003e\n\u003cli\u003eHaddaway, N.R.; Macura, B.; Whaley, P.; Pullin, A.S. ROSES RepOrting standards for Systematic Evidence Syntheses: Pro forma, flow-diagram and descriptive summary of the plan and conduct of environmental systematic reviews and systematic maps. Environ. Evid. 2018, 7, 7. [Google Scholar] [CrossRef]\u003c/li\u003e\n\u003cli\u003eMoher, D.; Liberati, A.; Tetzlaff, J.; Altman, D.G.; PRISMA Group. Preferred reporting items for systematic reviews and meta-analyses: The PRISMA statement. Ann. Intern. Med. 2009, 151, 264\u0026ndash;269. [Google Scholar] [CrossRef] [PubMed][Green Version]\u003c/li\u003e\n\u003cli\u003ePradhan, B. A comparative study on the predictive ability of the decision tree, support vector machine and neuro-fuzzy models in landslide susceptibility mapping using GIS. Comput. Geosci. 2013, 51, 350\u0026ndash;365. [Google Scholar] [CrossRef]\u003c/li\u003e\n\u003cli\u003eYilmaz, I. Landslide susceptibility mapping using frequency ratio, logistic regression, artificial neural networks and their comparison: A case study from Kat landslides (Tokat\u0026mdash;Turkey). Comput. Geosci. 2009, 35, 1125\u0026ndash;1138. [Google Scholar] [CrossRef]\u003c/li\u003e\n\u003cli\u003eYilmaz, I. Comparison of landslide susceptibility mapping methodolo- gies for Koyulhisar, Turkey: Conditional probability, logistic regression, artificial neural networks, and support vector machine. Environ. Earth Sci. 2010, 61, 821\u0026ndash;836. [Google Scholar] [CrossRef]\u003c/li\u003e\n\u003cli\u003eYoussef, A.M.; Pourghasemi, H.R.; Pourtaghi, Z.S.; Al-Katheeri, M.M. Landslide susceptibility mapping using random forest, boosted regres- sion tree, classification and regression tree, and general linear models and comparison of their performance at Wadi Tayyah Basin, Asir Region, Saudi Arabia. Landslides 2016, 13, 839\u0026ndash;856. [Google Scholar] [CrossRef]\u003c/li\u003e\n\u003cli\u003eYao, X.; Tham, L.; Dai, F. Landslide susceptibility mapping based on support vector machine: A case study on natural slopes of Hong Kong, China. Geomorphology 2008, 101, 572\u0026ndash;582. [Google Scholar] [CrossRef]\u003c/li\u003e\n\u003cli\u003e\u003cbr\u003eZare, M.; Pourghasemi, H.R.; Vafakhah, M.; Pradhan, B. Landslide susceptibility mapping at Vaz Watershed (Iran) using an artificial neural network model: A comparison between multilayer perceptron (MLP) and radial basic function (RBF) algorithms. Arab. J. Geosci. 2013, 6, 2873\u0026ndash;2888. [Google Scholar] [CrossRef]\u003c/li\u003e\n\u003cli\u003eCruden, D.M., 1991. A simple definition of a landslide. Bulletin of Engineering Geology and the Environment, 43(1), pp.27-29\u003c/li\u003e\n\u003cli\u003eParkash, S., 2011. Historical records of socio-economically significant landslides in India. J South Asia Disaster Studies, 4(2), pp.177-204.\u003c/li\u003e\n\u003cli\u003eJime\u0026acute;nez-Pera\u0026acute;lvarez, J.D., El Hamdouni, R., Palenzuela, J.A., Irigaray, C. and Chaco\u0026acute;n, J., 2017. Landslide-hazard mapping through multi- technique activity assessment: an example from the Betic Cordillera (southern Spain). Landslides, 14(6), pp.1975-1991\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":true,"highlight":"","institution":"Chandigarh University","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Landslide risk, Landslide identification, Machine learning, Deep learning, Big data, Convolutional neural networks","lastPublishedDoi":"10.21203/rs.3.rs-4632694/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-4632694/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eLandslides are natural disasters that can cause significant damage to the environment and pose a serious threat to human lives and infrastructure. Early detection and identification of potential landslide-prone areas are crucial for disaster mitigation and preparedness efforts. This abstract out- lines a comprehensive approach to landslide identification uti- lizing machine learning techniques. In recent years, machine learning has emerged as a powerful tool for analyzing geospatial data and predicting geological hazards such as landslides. This research leverages a diverse range of data sources, including remote sensing imagery, topographical maps, rainfall records, and geological data, to develop a robust landslide identification model. The key components of the proposed methodology involve data preprocessing, feature engineering, and the application of various machine learning algorithms. Remote sensing data, such as satellite imagery and LiDAR data, are used to extract valuable terrain features and land cover information. Rainfall data are incorporated to assess the influence of precipitation on landslide occurrence. Geological data contribute to the understanding of local geological conditions. Several machine learning algorithms, including but not limited to decision trees, support vector machines, and neural networks, are employed to create predictive models. These models are trained on historical landslide data and validated against real-world cases. Cross-validation techniques are applied to ensure the model\u0026rsquo;s robustness and generalization capabilities.\u003c/p\u003e","manuscriptTitle":"Predicting Landslides with Machine Learning: A Data-Driven Approach","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-06-26 14:46:42","doi":"10.21203/rs.3.rs-4632694/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"003d2450-9e7f-426c-9743-00bca2389dc0","owner":[],"postedDate":"June 26th, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":33683074,"name":"Artificial Intelligence and Machine Learning"}],"tags":[],"updatedAt":"2024-06-26T14:46:42+00:00","versionOfRecord":[],"versionCreatedAt":"2024-06-26 14:46:42","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-4632694","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-4632694","identity":"rs-4632694","version":["v1"]},"buildId":"qtupq5eGEP_6zYnWcrvyt","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}