Landslide susceptibility evaluation method based on isolated forest optimization

The Isolation Forest optimization method enhances landslide susceptibility assessments by selecting high-quality non-landslide samples based on similarity with landslide features, improving the accuracy and reliability of SVM and LR models.

CN119066573APending Publication Date: 2024-12-03GUILIN UNIV OF ELECTRONIC TECH +1
View PDF 3 Cites 0 Cited by

Patent Information

Application Number
CN202411094855.5
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-11
Publication Date
2024-12-03

AI Technical Summary

Technical Problem

Existing landslide susceptibility assessment methods suffer from low accuracy due to the use of low-quality non-landslide negative samples, which introduce noise into the model and lead to inaccurate predictions.

Method used

A landslide susceptibility evaluation method utilizing Isolation Forest optimization to learn the distribution characteristics of landslide data, select high-quality non-landslide negative samples based on similarity with landslide features, and integrate these samples with Support Vector Machine (SVM) and Logistic Regression (LR) models for training.

Benefits of technology

Significantly improves the accuracy and reliability of landslide susceptibility assessments by optimizing non-landslide negative samples, enhancing the performance of SVM and LR models, and reducing model overfitting.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119066573A_ABST
    Figure CN119066573A_ABST
Patent Text Reader

Abstract

The invention provides a landslide susceptibility evaluation method based on isolated forest optimization. The method comprises the following steps: 1, collecting environmental factors of landslide occurrence in a research area range, and constructing an environmental factor evaluation system in combination with historical landslide disaster points; 2, constructing a landslide positive sample data set based on historical landslide points and influence factors; step 3, inputting the landslide positive sample data set into an isolated forest (iForest for short) model for training, and extracting environmental attribute features of landslide disaster samples; 4, pre-selecting non-landslide points in the research area and outside the landslide point buffer area, constructing a pre-selected non-landslide negative sample, and preferably selecting the non-landslide negative sample by adopting the trained isolated forest model; and 5, constructing a comprehensive sample data set by using the optimized non-landslide negative samples and the landslide positive samples, respectively inputting the comprehensive sample data set into a support vector machine (SVM) and a logistic regression (LR) model for training, and verifying the effectiveness of the optimized non-landslide negative samples of the isolated forest. According to the method, the non-landslide negative sample is optimized through the isolated forest model, and training and prediction are performed in combination with the machine learning model, so that the problem of low model prediction precision caused by random selection of the non-landslide negative sample in a traditional method is solved, and the accuracy of a landslide susceptibility evaluation result is remarkably improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention belongs to the technical field of geological disaster prevention and control, and in particular relates to a landslide susceptibility evaluation method based on isolated forest optimization. Background Art

[0002] Landslide susceptibility assessment is generally considered the first step in assessing landslide hazards. When assessing landslide susceptibility based on machine learning, it is essentially a binary classification study, using the model to distinguish landslide points from non-landslide points to predict the probability of landslide disasters in other areas. The accuracy of landslide prediction results depends on the selection of landslide positive samples and non-landslide negative samples. Landslide positive samples can be generated through field surveys, landslide remote sensing interpretation, and landslide historical cataloging data. In contrast, non-landslide negative samples are less certain. In the past, researchers have mainly used the following methods to obtain non-landslide negative samples: random selection, creation of landslide buffers, and generation of low-slope rivers and terrain areas. However, these areas contain many unstable slopes and unidentified landslides. These methods rely on subjective guesswork or random selection and cannot make good use of existing landslide causes. Training landslide positive samples with low-quality non-landslide negative samples through a classification model will introduce noise into the model, reduce classification accuracy, and lead to inaccurate landslide susceptibility assessment results.

[0003] Isolation Forest (iForest) is a machine learning algorithm for anomaly detection. It is suitable for processing datasets with imbalanced labels and performs well in unsupervised learning, high-dimensional data, and diverse datasets. There are many anomaly detection algorithms, including methods based on clustering, reconstruction, statistics, and neural networks. Commonly used models include OCSVM, LOF, and iForest. These algorithms only require one type of sample data for training, and can effectively establish a feature distribution model for that type of sample, thereby determining whether the new input sample conforms to the feature distribution of that type of sample. They are widely used in various anomaly detection scenarios. Summary of the invention

[0004] In view of the above problems, the purpose of the present invention is to provide a landslide susceptibility assessment method based on isolation forest optimization to solve the problem of low model prediction accuracy caused by low quality of non-landslide negative samples and to improve the accuracy of landslide susceptibility assessment results.

[0005] In order to achieve the above-mentioned object and other related objects, the present invention provides a landslide susceptibility evaluation method based on isolated forest optimization, which comprises the following steps:

[0006] Step 1: According to the geomorphological, hydrological and geological environmental conditions of the study area, collect the environmental factors of landslides in the study area, analyze the development characteristics of landslide disasters in detail, determine the key environmental factors including elevation, slope, slope aspect, land use type and vegetation coverage, and build a systematic environmental factor evaluation system in combination with historical landslide disaster points.

[0007] Step 2: Preprocess the environmental factors and construct a landslide positive sample data set based on historical landslide points and environmental factors.

[0008] Furthermore, step 2 specifically includes:

[0009] 2-1 Preprocess the data of environmental factors, including reclassifying the layers of factors affecting the continuity of elevation, slope, terrain relief, surface cutting depth, terrain roughness, plane curvature and profile curvature, and then assigning values ​​to each sub-category;

[0010] 2-2 Assign values ​​to each subcategory of land use type and lithology discreteness influencing factors;

[0011] 2-3 Overlay the location information of historical landslide points with the processed environmental factor layer, and use the "Multi-value Extraction to Point" tool in ArcGIS software to extract the attribute values ​​of each environmental factor corresponding to each landslide point;

[0012] 2-4 Based on the data processed in steps 2-1, 2-2 and 2-3, a positive sample data set of landslides is constructed, which includes historical landslide points and their environmental factor attribute values.

[0013] Step 3: Input the constructed landslide positive sample data set into the isolation forest (iForest) model for training, so that the model can learn the distribution characteristics of landslide samples.

[0014] Furthermore, step three specifically includes:

[0015] 3-1 Input the constructed landslide positive sample data set into the isolation forest (iForest) model for training;

[0016] 3-2 Adjust the model parameters so that the isolation forest can effectively learn the distribution characteristics of landslide positive samples.

[0017] Furthermore, the core idea of ​​selecting non-landslide negative samples through isolation forest is that isolation forest constructs decision trees by randomly selecting features and feature values, which enables it to learn the normal distribution characteristics of landslide data when there is only landslide data, and then identify abnormal points that do not conform to this distribution. The specific execution process of isolation forest is:

[0018] (1) Randomly select n sample data X = {x1, ..., xn} from the data set as the training set for training the isolated tree (iTree);

[0019] (2) Randomly select the jth feature value q from the features of the training set, and randomly select a split value p between the maximum and minimum values ​​of the feature;

[0020] (3) Generate a hyperplane using the split value p in step (2) to divide the data space of the current node into two subspaces, and then place the samples with eigenvalues ​​less than the split value p in the left branch of the current node, and place the samples with eigenvalues ​​greater than or equal to p in the right branch of the current node;

[0021] (4) Recursively repeat steps (2) and (3) on the left and right branch nodes, and continuously construct new leaf nodes until there is only one data on the leaf node (no further cutting is possible) or the tree has grown to the set height;

[0022] (5) For each data point, calculate its path length h(x) in each isolated tree, that is, the distance from the root node to the leaf node;

[0023] (6) Take the average of the path lengths of all isolated trees to obtain the average path length E(h(x)), and then calculate the anomaly score. The calculation formula for the anomaly score s(x,n) is:

[0024]

[0025] Where c(n) is a normalization factor used to adjust the expected path length of the tree and is defined as:

[0026]

[0027] Here, H(i) is the i-th harmonic number, which is approximately ln(i)+0.5772156649 (Euler constant);

[0028] (7) Determine normality and abnormality. The value range of the abnormality score s(x,n) is between 0 and 1. When s(x,n) is close to 1, it means that the data point is easy to isolate and is an abnormal point, and the return value is -1; when s(x,n) is close to 0, it means that the data point is difficult to isolate and is a normal point, and the return value is 1.

[0029] Step 4: Pre-select non-landslide points within the study area and outside the landslide buffer zone, construct pre-selected non-landslide negative samples, and use the isolation forest model trained in step 3 to select the non-landslide negative samples. According to the distribution characteristics of the landslide positive samples learned during the training process, identify the data with low similarity to the landslide positive samples, and select the data with low similarity as the non-landslide negative samples, so as to select the non-landslide negative samples.

[0030] Furthermore, step four specifically includes:

[0031] 4-1 Use the "Create Random Points" tool in ArcGIS software to establish a 500-meter buffer zone with the landslide point as the center, and generate random points within the study area and outside the buffer zone. The number of random points generated is set to five times the number of landslide points, and the minimum distance between two random points is set to 300 meters;

[0032] 4-2 The random points generated in step 4-1 are used as pre-selected non-landslide points. The “Multi-value Extraction to Point” tool is used in ArcGIS software to extract the attribute values ​​of each environmental factor corresponding to each pre-selected non-landslide point to construct a pre-selected non-landslide negative sample;

[0033] 4-3 The pre-selected non-landslide negative samples described in step 4-2 are input into the isolation forest model trained in step 3 for prediction. According to their similarity with the landslide features learned during the model training process, these pre-selected non-landslide negative samples are assigned output results of -1 and 1, where -1 represents low similarity, and data with low similarity are selected as non-landslide negative samples.

[0034] Step 5: A comprehensive sample data set is constructed by selecting the non-landslide negative samples and the landslide positive samples, and they are input into the support vector machine (SVM) and logistic regression (LR) models for training respectively to verify the effectiveness of the isolation forest optimization of non-landslide negative samples.

[0035] Furthermore, step five specifically includes:

[0036] 5-1 Combine the non-landslide negative samples and landslide positive samples selected in step 4 to construct a comprehensive sample data set. The comprehensive sample data set includes the same number of landslide positive samples and non-landslide negative samples;

[0037] 5-2 Divide the comprehensive sample data set into training set and test set in a ratio of 7:3 for model training and evaluation;

[0038] 5-3 Input the training set data into the support vector machine (SVM) and logistic regression (LR) models for training;

[0039] 5-4 Use the test set data to evaluate the trained support vector machine (SVM) and logistic regression (LR) models, calculate the performance indicators of the models, and verify the effectiveness of the isolation forest optimization of non-landslide negative samples.

[0040] Compared with the prior art, the advantages of the present invention are:

[0041] Aiming at the problem that the traditional method of randomly selecting or creating a landslide buffer zone to select non-landslide negative samples leads to low model prediction accuracy, the present invention proposes a landslide susceptibility evaluation method based on isolation forest optimization. By optimizing non-landslide negative samples through isolation forest, and combining support vector machine (SVM) and logistic regression (LR) model training, the performance index of the model is significantly improved, and the effectiveness of isolation forest in selecting non-landslide negative samples is verified. The non-landslide negative samples optimized by isolation forest meet the demand of low susceptibility, effectively solve the problem of low quality of non-landslide negative samples in traditional methods, and significantly improve the accuracy and reliability of landslide susceptibility evaluation results. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to further illustrate the contents described in the present invention, the specific embodiments of the present invention are further described in detail below in conjunction with the accompanying drawings. It should be understood that these drawings are only used as typical examples and should not be regarded as limiting the scope of the present invention.

[0043] Figure 1 It is a general flow chart of a landslide susceptibility evaluation method based on isolated forest optimization according to the present invention.

[0044] Figure 2 It is a specific implementation flow chart of a landslide susceptibility evaluation method based on isolated forest optimization of the present invention.

[0045] Figure 3 It is a schematic diagram of ROC curves of various machine learning models in specific embodiments of the present invention. DETAILED DESCRIPTION

[0046] The following describes the embodiments of the present invention through specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the following specific embodiments are only used to explain this application.

[0047] The embodiment of the present invention provides a landslide susceptibility assessment method based on isolation forest optimization. The specific idea is to use isolation forest to learn the normal distribution characteristics of landslide data when there are only landslide data, and then identify abnormal points that do not conform to this distribution. Then, the pre-selected non-landslide negative samples are input into the trained isolation forest model. According to the similarity of landslide characteristics learned during the model training process, these pre-selected non-landslide negative samples are assigned output results of -1 and 1, where -1 represents low similarity, and data with low similarity are selected as non-landslide negative samples. Finally, they are respectively input into support vector machine (SVM) and logistic regression (LR) models for training to verify the effectiveness of isolation forest optimization of non-landslide negative samples.

[0048] like Figure 1 As shown, it is a general flow chart of a landslide susceptibility evaluation method based on isolated forest optimization. The specific implementation process of the present invention is as follows:

[0049] Step 1: According to the geomorphological, hydrological and geological environmental conditions of the study area, collect the environmental factors of landslides in the study area, analyze the development characteristics of landslide disasters in detail, determine the key environmental factors including elevation, slope, slope aspect, land use type and vegetation coverage, and build a systematic environmental factor evaluation system in combination with historical landslide disaster points.

[0050] Step 2: Preprocess the environmental factors and construct a landslide positive sample data set based on historical landslide points and environmental factors.

[0051] Furthermore, step 2 specifically includes:

[0052] 2-1 Preprocess the data of environmental factors, including reclassifying the layers of factors affecting the continuity of elevation, slope, terrain relief, surface cutting depth, terrain roughness, plane curvature and profile curvature, and then assigning values ​​to each sub-category;

[0053] 2-2 Assign values ​​to each subcategory of land use type and lithology discreteness influencing factors;

[0054] 2-3 Overlay the location information of historical landslide points with the processed environmental factor layer, and use the "Multi-value Extraction to Point" tool in ArcGIS software to extract the attribute values ​​of each environmental factor corresponding to each landslide point;

[0055] 2-4 Based on the data processed in steps 2-1, 2-2 and 2-3, a positive sample data set of landslides is constructed, which includes historical landslide points and their environmental factor attribute values.

[0056] Step 3: Input the constructed landslide positive sample data set into the isolation forest (iForest) model for training, so that the model can learn the distribution characteristics of landslide positive samples.

[0057] Furthermore, step three specifically includes:

[0058] 3-1 Input the constructed landslide positive sample data set into the isolation forest (iForest) model for training;

[0059] 3-2 Adjust the model parameters so that the isolation forest can effectively learn the distribution characteristics of landslide positive samples.

[0060] Furthermore, the isolation forest execution process is as follows:

[0061] (1) Randomly select n sample data X = {x1, ..., xn} from the data set as the training set for training the isolated tree (iTree);

[0062] (2) Randomly select the jth feature value q from the features of the training set, and randomly select a split value p between the maximum and minimum values ​​of the feature;

[0063] (3) Generate a hyperplane using the split value p in step (2) to divide the data space of the current node into two subspaces, and then place the samples with eigenvalues ​​less than the split value p in the left branch of the current node, and place the samples with eigenvalues ​​greater than or equal to p in the right branch of the current node;

[0064] (4) Recursively repeat steps (2) and (3) on the left and right branch nodes, and continuously construct new leaf nodes until there is only one data on the leaf node (no further cutting is possible) or the tree has grown to the set height;

[0065] (5) For each data point, calculate its path length h(x) in each isolated tree, that is, the distance from the root node to the leaf node;

[0066] (6) Take the average of the path lengths of all isolated trees to obtain the average path length E(h(x)), and then calculate the anomaly score. The calculation formula for the anomaly score s(x,n) is:

[0067]

[0068] Where c(n) is a normalization factor used to adjust the expected path length of the tree and is defined as:

[0069]

[0070] Here, H(i) is the i-th harmonic number, which is approximately ln(i)+0.5772156649 (Euler's constant).

[0071] (7) Determine normality and abnormality. The value range of the abnormality score s(x,n) is between 0 and 1. When s(x,n) is close to 1, it means that the data point is easy to isolate and is an abnormal point, and the return value is -1; when s(x,n) is close to 0, it means that the data point is difficult to isolate and is a normal point, and the return value is 1.

[0072] Step 4: Pre-select non-landslide points within the study area and outside the landslide buffer zone, construct pre-selected non-landslide negative samples, and use the isolation forest model trained in step 3 to select the non-landslide negative samples. According to the distribution characteristics of the landslide positive samples learned during the training process, identify the data with low similarity to the landslide positive samples, and select the data with low similarity as the non-landslide negative samples, so as to select the non-landslide negative samples.

[0073] Furthermore, step four specifically includes:

[0074] 4-1 Use the “Create Random Points” tool in ArcGIS software to establish a 500-meter buffer zone with the landslide point as the center, and generate random points within the study area and outside the buffer zone. The number of random points generated is set to five times the number of landslide points, and the minimum distance between two random points is set to 300 meters.

[0075] 4-2 The random points generated in step 4-1 are used as pre-selected non-landslide points. The “Multi-value Extraction to Point” tool is used in ArcGIS software to extract the attribute values ​​of each environmental factor corresponding to each pre-selected non-landslide point to construct a pre-selected non-landslide negative sample;

[0076] 4-3 The pre-selected non-landslide negative samples described in step 4-2 are input into the isolation forest model trained in step 3 for prediction. According to their similarity with the landslide features learned during the model training process, these pre-selected non-landslide negative samples are assigned output results of -1 and 1, where -1 represents low similarity, and data with low similarity are selected as non-landslide negative samples.

[0077] Step 5: A comprehensive sample data set is constructed by selecting the non-landslide negative samples and the landslide positive samples, and they are input into the support vector machine (SVM) and logistic regression (LR) models for training respectively to verify the effectiveness of the isolation forest optimization of non-landslide negative samples.

[0078] Furthermore, step five specifically includes:

[0079] 5-1 Combine the non-landslide negative samples and landslide positive samples selected in step 4 to construct a comprehensive sample data set. The comprehensive sample data set includes the same number of landslide positive samples and non-landslide negative samples;

[0080] 5-2 Divide the comprehensive sample data set into training set and test set in a ratio of 7:3 for model training and evaluation;

[0081] 5-3 Input the training set data into the support vector machine (SVM) and logistic regression (LR) models for training and optimize the model parameters;

[0082] Specifically, in step 5-3, the model parameter tuning is as follows:

[0083] (1) Construct a logistic regression (LR) model, define candidate values ​​for the penalty function "penalty" and the regularization parameter "c", and use cross-validation to evaluate the effects of different parameter combinations. There are two options for the penalty function "penalty". "l1" represents L1 regularization (Lasso), which causes the weights of some features to become zero, thereby achieving feature selection; "l2" represents L2 regularization (Ridge), which causes the weights to tend to zero but not completely become zero, thereby avoiding overfitting. The regularization parameter "c" controls the regularization strength. The larger the value, the smaller the regularization strength; the smaller the value, the greater the regularization strength, thereby controlling the complexity of the model and preventing overfitting.

[0084] Specifically, all parameter combinations are looped through, the AUC value of each combination in cross validation is evaluated, the hyperparameter combination with the highest AUC value is selected, and then the best hyperparameter combination is used to train the logistic regression (LR) model.

[0085] (2) Construct a support vector machine (SVM) model, define candidate values ​​for the kernel function coefficient "gamma" and the regularization parameter "c", and use cross-validation to evaluate the effects of different parameter combinations. The kernel function coefficient "gamma" controls the influence range of a single training sample. The larger the value, the smaller the influence range of each sample, and the more complex the model; the smaller the value, the larger the influence range of the sample, and the simpler the model. The regularization parameter "c" controls the degree of penalty for misclassification. The larger the value, the lower the tolerance of the model to misclassification, and the more complex the model; the smaller the value, the higher the tolerance of the model to misclassification, and the simpler the model.

[0086] Specifically, all parameter combinations are looped through, the AUC value of each combination in cross validation is evaluated, the hyperparameter combination with the highest AUC value is selected, and then the support vector machine (SVM) model is trained using the best hyperparameter combination.

[0087] 5-4 Use the test set data to evaluate the trained classification model, calculate the model performance indicators, and compare the results of the SVM and LR models with and without the isolation forest optimization of non-landslide negative samples to verify the effectiveness of the isolation forest optimization of non-landslide negative samples.

[0088] Specifically, by plotting the relationship between sensitivity (i.e., true positive rate, indicating the ability of the model to correctly identify positive samples) on the y-axis and specificity (i.e., false positive rate, indicating the proportion of negative samples incorrectly identified as positive samples by the model) on the x-axis, a receiver operating characteristic (ROC) curve is generated, and its area under the line (AUC) is calculated to evaluate the performance of the prediction model. The closer the AUC value is to 1, the better the overall classification performance of the model. Figure 3The ROC curves of the models of the present invention are shown in Figure 2. In addition, the accuracy of each model is evaluated by combining the overall accuracy (OA), precision, recall, F1 score and Matthews correlation coefficient (MCC) to verify the classification effect of the model. The higher the values ​​of these indicators, the better the classification performance of the model.

[0089] Specifically, the calculation formulas for various performance indicators are as follows:

[0090]

[0091] Specifically, TP (True Positive) and FP (False Positive) are the number of samples correctly classified and misclassified as landslides, and TN (True Negative) and FN (False Negative) are the number of samples correctly classified and misclassified as non-landslides.

[0092] Specifically, Table 1 shows the accuracy statistics of each model. The data in the table show that after the non-landslide samples are selected by the isolation forest model (iForest), the performance of the SVM and LR models are significantly improved. iForest-SVM and iForest-LR are better than the SVM and LR models without iForest in terms of OA, AUC, MCC, Precision, Recall, and F-score. iForest-SVM improves the above indicators by 0.119, 0.103, 0.238, 0.130, 0.118, and 0.124 respectively compared with SVM; iForest-LR improves the above indicators by 0.114, 0.108, 0.227, 0.102, 0.144, and 0.124 respectively compared with LR. These data show that the iForest model can effectively improve the quality of non-landslide samples and significantly improve the performance of the classification model, indicating the effectiveness of the isolation forest in selecting non-landslide samples in improving the prediction accuracy of the model.

[0093] Table 1 Model accuracy statistics

[0094] Model OA AUC MCC Precision Recall F-score iForest-SVM 0.890 0.959 0.782 0.926 0.847 0.885 iForest-LR 0.856 0.927 0.712 0.868 0.839 0.853 SVM 0.771 0.856 0.544 0.796 0.729 0.761 LR 0.742 0.819 0.485 0.766 0.695 0.729

[0095] Compared with the prior art solutions, the present invention has the following beneficial effects:

[0096] The present invention proposes a landslide susceptibility evaluation method based on isolation forest optimization, which significantly improves the performance indicators of support vector machine (SVM) and logistic regression (LR) models, including overall accuracy, AUC value, precision, recall rate, F1 score and Matthews correlation coefficient, verifies the effectiveness of isolation forest in selecting non-landslide negative samples, and solves the problem of low quality of non-landslide negative samples in traditional methods. The non-landslide negative samples optimized by isolation forest meet the requirements of low susceptibility and significantly improve the accuracy and reliability of landslide susceptibility evaluation results. By adjusting the regularization parameters, model overfitting is avoided and the generalization ability of the model is improved. Compared with the traditional method of randomly selecting or creating a buffer zone to select non-landslide negative samples, the present invention provides a more efficient and scientific non-landslide negative sample selection method, which improves the prediction effect of the entire model.

[0097] The above embodiments are merely illustrative of the principles and effects of the present invention, and are not intended to limit the present invention. By referring to the detailed description of the embodiments, any person skilled in the art may modify or change the above embodiments without violating the spirit and scope of the present invention. Therefore, all equivalent modifications or changes made by a person of ordinary skill in the art without departing from the spirit and technical ideas disclosed by the present invention shall still be covered by the claims of the present invention.

Claims

1. A landslide susceptibility assessment method based on isolated forest optimization, It is characterized in that The method comprises the following steps: Step 1: According to the geomorphology, hydrology and geological environment conditions of the study area, collect the environmental factors of landslides in the study area, analyze the development characteristics of landslide disasters in detail, determine the key environmental factors including elevation, slope, slope direction, land use type and vegetation coverage, and build a systematic environmental factor evaluation system in combination with historical landslide disaster points; Step 2: Preprocess the environmental factors and construct a landslide positive sample data set based on historical landslide points and environmental factors; Step 3: Input the constructed landslide positive sample data set into the Isolation Forest (iForest) model for training, so that the model can learn the distribution characteristics of landslide positive samples; Step 4: Preselect non-landslide points within the study area and outside the landslide buffer zone, construct preselected non-landslide negative samples, use the isolation forest model trained in step 3 to select the non-landslide negative samples, and identify the data with low similarity to the landslide positive samples according to the distribution characteristics of the landslide positive samples learned during its training process, and select the data with low similarity as the non-landslide negative samples, so as to select the non-landslide negative samples; Step 5: A comprehensive sample data set is constructed by selecting the non-landslide negative samples and the landslide positive samples, and they are input into the support vector machine (SVM) and logistic regression (LR) models for training respectively to verify the effectiveness of the isolation forest optimization of non-landslide negative samples.

2. The landslide susceptibility assessment method based on isolated forest optimization according to claim 1, It is characterized in that The step 2 is specifically as follows: 2-1 Preprocess the data of environmental factors, including reclassifying the layers of factors affecting the continuity of elevation, slope, terrain relief, surface cutting depth, terrain roughness, plane curvature and profile curvature, and then assigning values ​​to each sub-category; 2-2 Assign values ​​to each subcategory of land use type and lithology discreteness influencing factors; 2-3 Overlay the location information of historical landslide points with the processed environmental factor layer, and use the "Multi-value Extraction to Point" tool in ArcGIS software to extract the attribute values ​​of each environmental factor corresponding to each landslide point; 2-4 Based on the data processed in steps 2-1, 2-2 and 2-3, a positive sample data set of landslides is constructed, which includes historical landslide points and their environmental factor attribute values.

3. The landslide susceptibility assessment method based on isolated forest optimization according to claim 1, It is characterized in that The step three is specifically as follows: 3-1 Input the constructed landslide positive sample data set into the isolation forest (iForest) model for training; 3-2 Adjust the model parameters so that the isolation forest can effectively learn the distribution characteristics of landslide positive samples. Furthermore, the isolation forest execution process is as follows: (1) Randomly select n sample data X = {x1, ..., xn} from the data set as the training set for training the isolated tree (iTree); (2) Randomly select the jth feature value q from the features of the training set, and randomly select a split value p between the maximum and minimum values ​​of the feature; (3) Generate a hyperplane using the split value p in step (2) to divide the data space of the current node into two subspaces, and then place the samples with eigenvalues ​​less than the split value p in the left branch of the current node, and place the samples with eigenvalues ​​greater than or equal to p in the right branch of the current node; (4) Recursively repeat steps (2) and (3) on the left and right branch nodes, and continuously construct new leaf nodes until there is only one data on the leaf node (no further cutting is possible) or the tree has grown to the set height; (5) For each data point, calculate its path length h(x) in each isolated tree, that is, the distance from the root node to the leaf node; (6) Take the average of the path lengths of all isolated trees to obtain the average path length E(h(x)), and then calculate the anomaly score. The calculation formula for the anomaly score s(x,n) is: Where c(n) is a normalization factor used to adjust the expected path length of the tree and is defined as: Here, H(i) is the i-th harmonic number, which is approximately ln(i)+0.5772156649 (Euler constant); (7) Determine normality and abnormality. The value range of the abnormality score s(x,n) is between 0 and 1. When s(x,n) is close to 1, it means that the data point is easy to isolate and is an abnormal point, and the return value is -1; when s(x,n) is close to 0, it means that the data point is difficult to isolate and is a normal point, and the return value is 1.

4. The landslide susceptibility assessment method based on isolated forest optimization according to claim 1, It is characterized in that The step 4 is specifically as follows: 4-1 Use the "Create Random Points" tool in ArcGIS software to establish a 500-meter buffer zone with the landslide point as the center, and generate random points within the study area and outside the buffer zone. The number of random points generated is set to five times the number of landslide points, and the minimum distance between two random points is set to 300 meters; 4-2 The random points generated in step 4-1 are used as pre-selected non-landslide points. The "Multi-value Extraction to Point" tool is used in ArcGIS software to extract the attribute values ​​of each environmental factor corresponding to each pre-selected non-landslide point to construct a pre-selected non-landslide negative sample; 4-3 The pre-selected non-landslide negative samples described in step 4-2 are input into the isolation forest model trained in step 3 for prediction. According to their similarity with the landslide features learned during the model training process, these pre-selected non-landslide negative samples are assigned output results of -1 and 1, where -1 represents low similarity, and data with low similarity are selected as non-landslide negative samples.

5. The landslide susceptibility assessment method based on isolated forest optimization according to claim 1, It is characterized in that The step five is specifically as follows: 5-1 The non-landslide negative samples and the landslide positive samples selected in step 4 are merged to construct a comprehensive sample data set, which includes the same number of landslide positive samples and non-landslide negative samples; 5-2 Divide the comprehensive sample data set into training set and test set in a ratio of 7:3 for model training and evaluation; 5-3 Input the training set data into the support vector machine (SVM) and logistic regression (LR) models for training; 5-4 Use the test set data to evaluate the trained support vector machine (SVM) and logistic regression (LR) models, calculate the performance indicators of the models, and verify the effectiveness of the isolation forest optimization of non-landslide negative samples.

Citation Information

Patent Citations

  • Consumption credit fraud behavior detection method and system based on isolated forest

    CN111833172A

  • Fraudulent user detection method and device, medium and electronic equipment

    CN114298176A

  • Slope unit landslide susceptibility evaluation method based on optimized negative sample selection

    CN118332434A