Figure 2 - uploaded by Van-Manh Pham
Content may be subject to copyright.
Source publication
This study aims at investigating the balance between exploration and exploitation search capability of a newly developed Salp swarm optimization algorithm (SSA) for fine-tuning parameters of a three-hidden-layer neural network. The landslide study was selected as a thematic application, and a mountainous area of Vietnam was chosen as a case study....
Contexts in source publication
Context 1
... previously published works have verified the potential replacement of gradient descend by metaheuristic algorithms in fine-tuning parameters of the network. This study investigated the potential use of a newly developed optimizer SSA to search for optimal weights of a MLNN ( Figure 2). From the trial and error process, the structure of this network was determined to have one input layer, three hidden layers with 10, 9, 8 nodes (neurons), respectively, and one output layer. ...
Context 2
... network produces a susceptible value P i between [0 1] range which is used with the observed value O i to formulate the objective function (Root mean square error -RMSE) for the SSA algorithm. The brief description is shown in (Figure 2) SSA is used as the optimizer for the proposed neural network. It is a bio-inspired algorithm to mimic the swarming behaviours of the salp chain and was mathematically investigated by the work of ( Mirjalili et al. 2017). ...
Citations
... The network parameter (Parameters) [84] is the sum of the parameter quantities of all convolutional layers and fully-connected layers in the network, which represents the size of the network model. ...
Infrared ship target segmentation is the important basis of infrared guided weapon in the sea-air context. Typically, accurate infrared ship target segmentation relies on a large number of pixel-level labels. However, it is difficult to obtain them. To this end, we present a method of Semi-supervised Infrared Ship Target Segmentation with Dual Branch (SeISTS-DB), which utilizes a small amount of labeled data and a large amount of unlabeled data to train model and improve segmentation performance. There are three main contributions. First, we design a target segmentation branch to generate the pseudo labels for unlabeled data. It consists of a dual learning network and a segmentation network. The dual learning network generates pseudo labels with weights for unlabeled data. The segmentation network is trained using both labeled data and unlabeled data with pseudo labels to achieve target segmentation of infrared ship, obtaining the preliminary segmentation results. Secondly, we introduce an error segmentation pixel correction branch, which contains a student network and a teacher network, to modify the pixel category error of the preliminary segmentation map. Finally, the outputs of the two branches are combined to obtain the final segmentation result. The SeISTS-DB is compared with other fully-supervised and semi-supervised methods on the infrared ship images dataset. Experimental results demonstrate that when the labeled data accounts for 1/8 of the training data, the mean Intersection over Union (mIou) is respectively improved by 15.35% and 6.19% at most. Besides, it is also compared with other methods on the public IRSTD-1k dataset, when the proportion of labeled images is 1/8, the mIoU is respectively improved by 11.76% at most compared to the state-of-the-art semi-supervised methods, demonstrating its effectiveness.
... Finally, the extrapolation problem is considered a major problem when using machine learning to predict the flood risk or estimate the flood depth. Several studies have tried different techniques to solve this problem such as augmenting training data in different geographic locations Nguyen et al. 2020). However, this method is very expensive and is not feasible in areas that are difficult to access and in regions with limited data. ...
Flood prediction is an important task, which helps local decision-makers in taking effective measures to reduce damage to the people and economy. Currently, most studies use machine learning to predict flooding in a given region; however, the extrapolation problem is considered a major challenge when using these techniques and is rarely studied. Therefore, this study will focus on an approach to resolve the extrapolation problem in flood depth prediction by integrating machine learning (XGBoost, Extra-Trees (EXT), CatBoost (CB), and light gradient boost machines (LightGBM)) and hydraulic modeling under MIKE FLOOD. The results show that the hydraulic model worked well in providing the flood depth data needed to build the machine learning model. Among the four proposed machine learning models, XGBoost was found to be the best at solving the extrapolation problem in the estimation of flood depth, followed by EXT, CB, and LightGBM. Quang Binh province was hit by floods with depths ranging from 0 to 3.2 m. Areas with high flood depths are concentrated along and downstream of the two major rivers (Gianh and Nhat Le – Kien Giang).
... Redundant factors can heighten model complexity, leading to diminished performance (Nguyen et al. 2020a). Although there isn't a universal guide for factor selection, our study identified 11 conditioning factors based on the geo-environmental conditions of the study area, and a comprehensive literature review. ...
Globally, coastal erosion significantly impacts the socio-economic conditions and infrastructure development of coastal
regions, with Vietnam facing considerable challenges due to its extensive coastline. This study focuses on developing innovative
hybrid machine learning models, namely BLWL and CGLWL, which combine Locally Weighted Learning (LWL) and
two optimization techniques, namely Bagging and Cascade Generalization, respectively. Quang Nam Province in Vietnam
consistently affected by coastal erosions, serves as the case study. For model development, a set of historical coastal erosions
and the affecting factors, such as magnitude of horizontal flow (sea currents), wave height, wave direction, distance to fault,
geology, river density, elevation, curvature, aspect, slope degree, and topographic wetness index were collected and used
for generation of the database. For the selection and prioritization of affecting coastal erosion factors, Correlation Attribute
Evaluation (CAE) method was used. Performance of the models was evaluated using standard statistical measures: Accuracy
Assessment (ACC), Sensitivity (SST), Specificity (SPF), Root Mean Squared Errors (RMSE), Kappa (K), Positive Predictive
Value (PPV), and Negative Predictive Value (NPV), and Area Under the ROC Curve (AUC). Results indicated that the BLWL
model (AUC: 0.978) was the best, followed by CGLWL (AUC: 0.968) and LWL (AUC: 0.963) models in accurately predicting
coastal erosion susceptible areas. Therefore, it can be concluded that BLWL is a promising tool for the development of
coastal erosion susceptibility maps, facilitating effective planning and management to mitigate the impact of coastal erosion.
Keywords Coastal erosion · Machine learning · Locally weighted learning · GIS · Vietnam
... With a third of its land area consisting of hills and mountains, Vietnam is one of the countries most affected. According to data from the General Department of Disaster Reduction (Ministry of Agriculture and Rural Development), between 2000 and 2015, Vietnam was affected by 250 flash floods and landslides, causing 779 deaths Nguyen et al., 2020). ...
... In many cases, authors select all factors available in the study area, and then use algorithms to assess the importance of each element, as well as their permutability. In this study, the RF method was used to determine the importance of factors (Chang et al., 2022;Nguyen et al., 2020;Pham et al., 2020). The results showed that topographic factors (slope, direction and elevation) were the most important in determining T A B L E 3 The performance of the eight models proposed. ...
Landslides lead to widespread devastation and significant loss of life in mountainous regions around the world. Susceptibility assessments can provide critical data to help decision‐makers, for example, local authorities and other organizations, mitigating the landslide risk, although the accuracy of existing studies needs to be improved. This study aims to assess landslide susceptibility in the Thua Thien Hue province of Vietnam using deep neural networks (DNNs) and swarm‐based optimization algorithms, namely Adam, stochastic gradient descent (SGD), Artificial Rabbits Optimization (ARO), Tuna Swarm Optimization (TSO), Sand Cat Swarm Optimization (SCSO), Honey Badger Algorithm (HBA), Marine Predators Algorithm (MPA) and Particle Swarm Optimization (PSO). The locations of 945 landslides occurring between 2012 and 2022, along with 14 conditioning factors, were used as input data to build the DNN and DNN‐hybrid models. The performance of the proposed models was evaluated using the statistical indices receiver operating characteristic curve, area under the curve (AUC), root mean square error, mean absolute error (MAE), R ² and accuracy. All proposed models had a high accuracy of prediction. The DNN‐MPA model had the highest AUC value (0.95), followed by DNN‐HBA (0.95), DNN‐ARO (0.95), DNN‐Adam (0.95), DNN‐SGD (0.95), DNN‐TSO (0.93), DNN‐PSO (0.9) and finally DNN‐SCSO (0.83). High‐precision models have identified that the majority of the western region of Thua Thien Hue province is very highly susceptible to landslides. Models like the aforementioned ones can support decision‐makers in updating large‐scale sustainable land‐use strategies.
... Previous to the application of the modeling algorithm, the identification of explanatory factors is relevant. It is based on literature review, expert knowledge, availability of information, and data exploration [20]. Some examples of data exploration techniques are Logistic Regression (LR) [21][22][23], heuristics based on expert opinion [24,25], Discriminant Analysis [26], Markov Chain [18], machine learning techniques [27], and Exploratory Factor Analysis (EFA). ...
... This study shows the importance of NDWI for the characterization of landslide occurrence. It is consistent with Zhang et al. [125], Maqsoom et al. [126], and Nguyen et al. [20]. The water infiltrates when soil moisture uptake exceeds its water-holding capacity, resulting in subsurface runoff. ...
Landslides are one of the natural phenomena with more negative impacts on landscape, natural resources, and human health worldwide. Andean geomorphology, urbanization, poverty, and inequality make it more vulnerable to landslides. This research focuses on understanding explanatory landslide factors and promoting quantitative susceptibility mapping. Both tasks supply valuable knowledge for the Andean region, focusing on territorial planning and risk management support. This work addresses the following questions using the province of Azuay-Ecuador as a study area: (i) How do EFA and LR assess the significance of landslide occurrence factors? (ii) Which are the most significant landslide occurrence factors for susceptibility analysis in an Andean context? (iii) What is the landslide susceptibility map for the study area? The methodological framework uses quantitative techniques to describe landslide behavior. EFA and LR models are based on a historical inventory of 665 records. Both identified NDVI, NDWI, altitude, fault density, road density, and PC2 as the most significant factors. The latter factor represents the standard deviation, maximum value of precipitation, and rainfall in the wet season (January, February, and March). The EFA model was built from 7 latent factors, which explained 55% of the accumulated variance, with a medium item complexity of 1.5, a RMSR of 0.02, and a TLI of 0.89. This technique also identified TWI, fault distance, plane curvature, and road distance as important factors. LR's model, with AIC of 964.63, residual deviance of 924.63, AUC of 0.92, accuracy of 0.84, and Kappa of 0.68, also shows statistical significance for slope, roads density, geology, and land cover factors. This research encompasses a time-series analysis of NDVI, NDWI, and precipitation, including vegetation and weather dynamism for landslide occurrence. Finally, this methodological framework replaces traditional qualitative models based on expert knowledge, for quantitative approaches for the study area and the Andean region.
... The value of NDVI is inversely proportional to the probability of the occurrence of natural hazards. In recent years in the study area, the forested area has rapidly diminished, leading to an increase in the number and intensity of natural hazards such as flooding and landslides (Nguyen, 2022c;Nguyen et al., 2020). The NDVI value in the study area ranged from −0.3 to 0.8. ...
... NDBI describes the density of infrastructure in a region and is a key factor in predicting natural hazard susceptibility because construction influences soil water permeability, flow velocity, and geomorphological structure (Nguyen, 2022c, Nguyen et al., 2020. The value of NDBI ranged from −0.41 to 0.7 in the study area. ...
Natural hazards constitute a diverse category and are unevenly distributed in time and space. This hinders predictive efforts, leading to significant impacts on human life and economies. Multi‐hazard prediction is vital for any natural hazard risk management plan. The main objective of this study was the development of a multi‐hazard susceptibility mapping framework, by combining two natural hazards—flooding and landslides—in the North Central region of Vietnam. This was accomplished using support vector machines, random forest, and AdaBoost. The input data consisted of 4591 flood points, 1315 landslide points, and 13 conditioning factors, split into training (70%), and testing (30%) datasets. The accuracy of the models' predictions was evaluated using the statistical indices root mean square error, area under curve (AUC), mean absolute error, and coefficient of determination. All proposed models were good at predicting multi‐hazard susceptibility, with AUC values over 0.95. Among them, the AUC value for the support vector machine model was 0.98 and 0.99 for landslide and flood, respectively. For the random forest model, these values were 0.98 and 0.98, and for AdaBoost, they were 0.99 and 0.99. The multi‐hazard maps were built by combining the landslide and flood susceptibility maps. The results showed that approximately 60% of the study area was affected by landslides, 30% by flood, and 8% by both hazards. These results illustrate how North Central is one of the regions of Vietnam that is most severely affected by natural hazards, particularly flooding, and landslides. The proposed models adapt to evaluate multi‐hazard susceptibility at different scales, although expert intervention is also required, to optimize the algorithms. Multi‐hazard maps can provide a valuable point of reference for decision makers in sustainable land‐use planning and infrastructure development in regions faced with multiple hazards, and to prevent and reduce more effectively the frequency of floods and landslides and their damage to human life and property.
... To reduce the problem's effects, several techniques were applied in this study, such as setting a dropout rate and limiting the search range. However, other methods can also be tried, such as collecting training data from different geographical locations (Nguyen et al. 2020). ...
Flood damage is becoming increasingly severe in the context of climate change and changes in land use. Assessing the effects of these changes on floods is important, to help decision-makers and local authorities understand the causes of worsening floods and propose appropriate measures. The objective of this study was to evaluate the effects of climate and land use change on flood susceptibility in Thua Thien Hue province, Vietnam, using machine learning techniques (support vector machine (SVM) and random forest (RF)) and remote sensing. The machine learning models used a flood inventory including 1,864 flood locations and 11 conditional factors in 2017 and 2021, as the input data. The predictive capacity of the proposed models was assessed using the area under the curve (AUC), the root mean square error (RMSE), and the mean absolute error (MAE). Both proposed models were successful, with AUC values exceeding 0.95 in predicting the effects of climate and land use change on flood susceptibility. The RF model, with AUC = 0.98, outperformed the SVM model (AUC = 0.97). The areas most susceptible to flooding increased between 2017 and 2021 due to increased built-up area.
HIGHLIGHTS
Machine learning algorithms were applied for flood susceptibility modeling.;
The RF model had the highest AUC value (0.98).;
The areas highly flood susceptibility increased between 2017 and 2021.;
... RMSE and MAE are two popular statistical indices for analyzing landslide susceptibility model performance. They measure the errors between the prediction value and the observation value (Pham et al. 2019b;Nguyen et al. 2020;Pham et al. 2020), as in following equation: ...
Understanding the negative effects of climate change and changes to land use/land cover on natural hazards is an important feature of sustainable development worldwide, as these phenomena are inextricably linked with natural hazards such as landslides. The contribution of this study is an attempt to develop a state-of-the-art method to assess the effects of climate change and changes in land use/land cover on landslide susceptibility in the Tra Khuc river basin in Vietnam. The method is based on machine learning and remote sensing algorithms, namely radial basis function neural networks–search and rescue optimization (RBFNN–SARO), radial basis function neural network–queuing search algorithm (RBFNN–QSA), radial basis function neural network–life choice-based optimizer (RBFNN–LCBO), radial basis function neural network–dragonfly optimization (RBFNN–DO). All proposed models performed well, with AUC value of >0.9. The RBFNN–QSA model performed best, with an AUC value of 0.98, followed by RBFNN–SARO (AUC = 0.97), RBFNN–LCBO (AUC = 0.95), RBFNN–DO (AUC = 0.93), and support vector machine (SVM; AUC = 0.92). The results show that both climate and land use/land cover change greatly in the future: Precipitation increases 18% by 2030 and 25.1% by 2050; the total production forest, protected forest and built-up area change considerably between 2010 and 2050. These changes influence landslide susceptibility: The area of high and very high landslide susceptibility decrease by approximately 100 and 300 km² respectively in the study area from 2010 to 2050. The findings of this study can support decision-makers in formulating appropriate strategies to reduce damage from landslides, such as limiting construction in areas where future landslides are predicted. Although this study applies to a particular region of Vietnam, the findings can be applied in other mountainous regions around the world.
... For the first type, there are growing studies on the uses of optimization algorithms in searching for optimal parameters of classifiers in a broad range of applications, such as land cover classification (Q.-T. Bui et al., 2021;, landslide risk detection (Achour et al., 2017(Achour et al., , 2021Achour & Pourghasemi, 2020;Dou et al., 2015;Nguyen et al., 2020;Pham et al., 2019), forest fire analysis (Bui, 2019;Huu Duy, 2022), community disease assessment , flood susceptibility mapping (D. T. Bui et al., 2016Ngo et al., 2018;Nguyen et al., 2021;Razavi Termeh et al., 2018;Sachdeva et al., 2017). It could be seen from these works that all parameters are tunable, and they are updated during the training process . ...
Properly choosing hyper‐parameters improves machine learning models' performance and reduces training time and resource requirements. In this study, we investigated the uses of the Bayesian optimization algorithm for hyper‐parameter searches of two classifiers, namely LightGBM and XGBoost. The models were verified with a dataset from Vietnam, including historical flood locations from satellite images and survey data, and 11 features from three groups, namely physical, hydrological, and human‐related factors. The models' performance was evaluated using Area under Receiver Operating Characteristic curves (AUC‐ROC). Several strategies were applied to avoid over‐fitting, and the results show that two tuned Gradient boosters reached considerably high AUC values (approximately 0.98) compared with the previous study with a similar dataset. The model interpretation was also implemented using the Shapley (SHAP) values to understand better how models work and the interactions between features. The search for optimal hyper‐parameters is worth investigating in the future, particularly when there is growing work for novel optimization algorithms. The verification of such an approach is scientifically sound, and the models can be used as an alternative solution for natural hazard analysis in countries prone to hazards.
... A professional approach to slope instability assessment, disaster management, and mitigation is regarded as necessitating the use of landslide susceptibility mapping (LSM) [6,7]. Establishing the interdependence of past landslide incidents and causative factors-variables that are anticipated to have an impact on slide occurrences in a region-is a difficult procedure known as susceptibility mapping [8]. The creation of a landslide inventory, the identification of landslide causal factors (LCFs), the geographical correlation of these factors using a modelling methodology, and the model substantiation are some of the processes in landslide vulnerability study. ...
