Landslide susceptibility prediction (LSP) has been widely and effectively implemented by machine learning (ML) models based on remote sensing (RS) images and Geographic Information System (GIS). However, comparisons of the applications of ML models for LSP from the perspectives of supervised machine learning (SML) and unsupervised machine learning (USML) have not been explored. Hence, this study aims to compare the LSP performance of these SML and USML models, thus further to explore the advantages and disadvantages of these ML models and to realize a more accurate and reliable LSP result. Two representative SML models (support vector machine (SVM) and CHi-squared Automatic Interaction Detection (CHAID)) and two representative USML models (K-means and Kohonen models) are respectively used to scientifically predict the landslide susceptibility indexes, and then these prediction results are discussed. Ningdu County with 446 recorded landslides obtained through field investigations is introduced as case study. A total of 12 conditioning factors are obtained through procession of Landsat TM 8 images and high-resolution aerial images, topographical and hydrological spatial analysis of Digital Elevation Modeling in GIS software, and government reports. The area value under the curve of receiver operating features (AUC) is applied for evaluating the prediction accuracy of SML models, and the frequency ratio (FR) accuracy is then introduced to compare the remarkable prediction performance differences between SML and USML models. Overall, the receiver operation curve (ROC) results show that the AUC of the SVM is 0.892 and is slightly greater than the AUC of the CHAID model (0.872). The FR accuracy results show that the SVM model has the highest accuracy for LSP (77.80%), followed by the CHAID model (74.50%), the Kohonen model (72.8%) and the K-means model (69.7%), which indicates that the SML models can reach considerably better prediction capability than the USML models. It can be concluded that selecting recorded landslides as prior knowledge to train and test the LSP models is the key reason for the higher prediction accuracy of the SML models, while the lack of a priori knowledge and target guidance is an important reason for the low LSP accuracy of the USML models. Nevertheless, the USML models can also be used to implement LSP due to their advantages of efficient modeling processes, dimensionality reduction and strong scalability.

Landslides are one type of serious geological hazard which cause immense losses of local life and property. Landslide susceptibility prediction (LSP) can be used to determine the spatial probability of landslide occurrence in a certain area. It is important to implement LSP for landslide hazard prevention and reduction. This study developed a particle-swarm-optimized multilayer perceptron (PSO-MLP) model for LSP implementation to overcome the drawbacks of the conventional gradient descent algorithm and to determine the optimal structural parameters of MLP. Shicheng County in Jiangxi Province of China was used as the study area. In total, 369 landslides, randomly selected non-landslides, and 14 landslide-related predisposing factors were used to train and test the present PSO-MLP model and three other comparative models (an MLP-only model with the gradient descent algorithm, a back-propagation neural network (BPNN), and an information value (IV) model). The results showed that the PSO-MLP model had the most accurate prediction performance (area under the receiver operating characteristic curve (AUC) of 0.822 and frequency ratio (FR) accuracy of 0.856) compared with the MLP-only (0.791 and 0.829), BPNN (0.800 and 0.840), and IV (0.788 and 0.824) models. It can be concluded that the proposed PSO-MLP model addresses the drawbacks of the MLP-only model well and performs better than conventional artificial neural networks (ANNs) and statistical models. The spatial probability distribution law of landslide occurrence in Shicheng County was well revealed by the landslide susceptibility map produced using the PSO-MLP model. Furthermore, the present PSO-MLP model may have higher prediction and classification performances in some other fields compared with conventional ANNs and statistical models.

Landslide susceptibility prediction (LSP) modeling is an important and challenging problem. Landslide features are generally uncorrelated or nonlinearly correlated, resulting in limited LSP performance when leveraging conventional machine learning models. In this study, a deep-learning-based model using the long short-term memory (LSTM) recurrent neural network and conditional random field (CRF) in cascade-parallel form was proposed for making LSPs based on remote sensing (RS) images and a geographic information system (GIS). The RS images are the main data sources of landslide-related environmental factors, and a GIS is used to analyze, store, and display spatial big data. The cascade-parallel LSTM-CRF consists of frequency ratio values of environmental factors in the input layers, cascade-parallel LSTM for feature extraction in the hidden layers, and cascade-parallel full connection for classification and CRF for landslide/non-landslide state modeling in the output layers. The cascade-parallel form of LSTM can extract features from different layers and merge them into concrete features. The CRF is used to calculate the energy relationship between two grid points, and the extracted features are further smoothed and optimized. As a case study, the cascade-parallel LSTM-CRF was applied to Shicheng County of Jiangxi Province in China. A total of 2709 landslide grid cells were recorded and 2709 non-landslide grid cells were randomly selected from the study area. The results show that, compared with existing main traditional machine learning algorithms, such as multilayer perception, logistic regression, and decision tree, the proposed cascade-parallel LSTM-CRF had a higher landslide prediction rate (positive predictive rate: 72.44%, negative predictive rate: 80%, total predictive rate: 75.67%). In conclusion, the proposed cascade-parallel LSTM-CRF is a novel data-driven deep learning model that overcomes the limitations of traditional machine learning algorithms and achieves promising results for making LSPs.

Landslides are affected not only by their own environmental factors, but also by the neighborhood environmental factors and the landslide clustering effect, which are represented as the neighborhood characteristics of modelling spatial datasets in landslide susceptibility prediction (LSP). This study aims to innovatively explore the neighborhood characteristics of landslide spatial datasets for reducing the LSP uncertainty. Neighborhood environmental factors were acquired and managed by remote sensing (RS) and the geographic information system (GIS), then used to represent the influence of landslide neighborhood environmental factors. The landslide aggregation index (LAI) was proposed to represent the landslide clustering effect in GIS. Taking Chongyi County, China, as example, and using the hydrological slope unit as the mapping unit, 12 environmental factors including elevation, slope, aspect, profile curvature, plan curvature, topographic relief, lithology, gully density, annual average rainfall, NDVI, NDBI, and road density were selected. Next, the support vector machine (SVM) and random forest (RF) were selected to perform LSP considering the neighborhood characteristics of landslide spatial datasets based on hydrologic slope units. Meanwhile, a grid-based model was also established for comparison. Finally, the LSP uncertainties were analyzed from the prediction accuracy and the distribution patterns of landslide susceptibility indexes (LSIs). Results showed that the improved frequency ratio method using LAI and neighborhood environmental factors can effectively ensure the LSP accuracy, and it was significantly higher than the LSP results without considering the neighborhood conditions. Furthermore, the Wilcoxon rank test in nonparametric test indicates that the neighborhood characteristics of spatial datasets had a great positive influence on the LSP performance.

Soil erosion (SE) provides slide mass sources for landslide formation, and reflects long-term rainfall erosion destruction of landslides. Therefore, it is possible to obtain more reliable landslide susceptibility prediction results by introducing SE as a geology and hydrology-related predisposing factor. The Ningdu County of China is taken as a research area. Firstly, 446 landslides are obtained through government disaster survey reports. Secondly, the SE amount in Ningdu County is calculated and nine other conventional predisposing factors are obtained under both 30 m and 60 m grid resolutions to determine the effects of SE on landslide susceptibility prediction. Thirdly, four types of machine-learning predictors with 30 m and 60 m grid resolutions—C5.0 decision tree (C5.0 DT), logistic regression (LR), multilayer perceptron (MLP) and support vector machine (SVM)—are applied to construct the landslide susceptibility prediction models considering the SE factor as SE-C5.0 DT, SE-LR, SE-MLP and SE-SVM models; C5.0 DT, LR, MLP and SVM models with no SE are also used for comparisons. Finally, the area under receiver operating feature curve is used to verify the prediction accuracy of these models, and the relative importance of all the 10 predisposing factors is ranked. The results indicate that: (1) SE factor plays the most important role in landslide susceptibility prediction among all 10 predisposing factors under both 30 m and 60 m resolutions; (2) the SE-based models have more accurate landslide susceptibility prediction than the single models with no SE factor; (3) all the models with 30 m resolutions have higher landslide susceptibility prediction accuracy than those with 60 m resolutions; and (4) the C5.0 DT and SVM models show higher landslide susceptibility prediction performance than the MLP and LR models.