Machine learning-based techniques for land subsidence simulation in an urban area

Evaluating Machine Learning-Based Approaches in Land Subsidence Susceptibility Mapping

“…To prepare the input variables, Sentinel-1 satellite images of the study areas for 2015-2017 were obtained, and the locations and dimensions of several subsidence points in the areas were extracted using image analysis and by referring to technical reports (obtained from the Geological Survey and Mineral Exploration Institute) and previous studies [16][17][18][19][20][21][22][23][24][25][26][27][28][29][30][31]. The geographic location of the occurrence and the non-occurrence LS points (provided by the Geological Survey and Mineral Exploration Institute, the abovementioned studies, and RS-GIS-based methods) are used to obtain LS maps.…”

“…They reported that MARS outperformed other MLAs in the study area. Liu et al [20] addressed LS in urban planning and infrastructure management by using two machine learning models, including the extreme gradient boosting regressor (XGBR) and long short-term memory (LSTM). They identified groundwater level (GWL) and building concentration (BC) as key factors influencing LS.…”

“…To achieve these aims, we scrutinized the performance of six classification MLAs (classification and regression tree (CART), Bayesian linear regression (BLR), SVM, boosted BRT, RF, and logistic regression (LogR)) and one regression method (multiple linear regression (MLR)) in identifying LS-prone areas and predicting the magnitude of LS. We applied the methods to two study areas, Semnan Plain and Kashmar Plain in Iran, both of which have experienced severe LS in recent decades [16][17][18][19][20][21][22][23][24], using nine input variables: distance from the river, distance from the fault, groundwater drawdown, slope, aspect, land use, lithology, topographic wetness index (TWI), and normalized difference vegetation index (NDVI).…”

“…These seven MLAs were selected for analysis based on their performance merits and previous studies [16][17][18][19][20][21][22][23][24][25][26][27]. The CART approach can handle missing values in the training dataset and is insensitive to outliers since it uses surrogates [28].…”

Advanced Modulation Formats for 400 Gbps Optical Networks and AI-Based Format Recognition

“…Neural network technology utilizes deep neural networks with multiple nonlinear layers to learn and extract feature information from signals, without relying on manual experience to extract signal features [ 17 ]. MFI methods based on neural network technology have the advantages of requiring less prior knowledge, strong feature extraction capabilities, and high recognition accuracy, which have received widespread attention and research from scientific researchers [ 18 , 19 , 20 ].…”