Vehicle insurance anomaly detection model construction method and device, equipment and storage medium

Abstract

The invention relates to a method for constructing a vehicle insurance anomaly detection model, and the method comprises the steps: obtaining vehicle insurance policy information, obtaining vehicle driving track data corresponding to the vehicle insurance policy information, and marking the vehicle driving track data as normal or abnormal. And claim settlement features of the vehicle insurance policy information are extracted, trajectory features are extracted according to normal or abnormal labels of the vehicle driving trajectory data, and a vehicle insurance anomaly detection model is constructed by using an isolated forest algorithm according to the claim settlement features and the trajectory features. According to the method, a plurality of randomized binary trees are constructed through the isolated forest algorithm, abnormal data points can be effectively identified, the detection accuracy of abnormal data in vehicle insurance claim settlement is remarkably improved, processing and analysis of large-scale data can be completed in a short time, and the method is suitable for detection requirements of mass claim settlement data in actual services.

Description

Technical Field

[0001] The present invention relates to the field of intelligent detection, and in particular to a method, device, equipment and storage medium for constructing a vehicle insurance anomaly detection model. Background Art

[0002] With the increase in the number of cars, the auto insurance business has developed rapidly. However, there are a lot of frauds in the process of auto insurance claims, such as false reporting of accidents, forged evidence, exaggeration of losses, etc. These frauds not only increase the financial burden of insurance companies, but also disrupt the normal order of the insurance market. Therefore, how to effectively identify and prevent auto insurance fraud has become an urgent problem that the insurance industry needs to solve.

[0003] Traditional auto insurance fraud detection methods mainly rely on manual review and simple rule-based judgment. The shortcomings of these methods are:

[0004] Inefficiency: Manual review requires a lot of manpower and time, and it is difficult to cope with a large number of claims cases.

[0005] Limited accuracy: Rule-based detection methods can only capture known fraud patterns and are often helpless against new or mutated fraud behaviors.

[0006] Strong subjectivity: Manual review is subject to subjective judgment and is easily affected by the reviewer’s experience and emotions.

[0007] In recent years, with the development of big data and artificial intelligence technology, data-driven fraud detection methods have gradually gained attention. Through machine learning algorithms, patterns of fraudulent behavior can be automatically learned and discovered from massive historical data, thereby improving the efficiency and accuracy of detection. As an effective anomaly detection algorithm, the isolation forest algorithm has been widely used in fraud detection due to its high computational efficiency and superior detection performance.

[0008] The Isolation Forest Algorithm is an unsupervised learning method based on a tree model. It measures the isolation of data points by constructing multiple randomized binary trees. The core idea of ​​the Isolation Forest Algorithm is that outliers are usually isolated at a lower level in the tree structure, that is, the path is shorter. Therefore, the path length of the data point in the tree can be calculated to determine whether it is an outlier. Summary of the invention

[0009] The main purpose of the present invention is to provide a method, device, equipment and storage medium for constructing a motor vehicle insurance anomaly detection model, aiming to solve the technical problem of low detection accuracy of existing motor vehicle insurance fraud risk detection models.

[0010] To achieve the above object, the present invention provides a method for constructing a vehicle insurance anomaly detection model, and the method for constructing a vehicle insurance anomaly detection model comprises the following steps:

[0011] Obtaining automobile insurance policy information, obtaining vehicle driving trajectory data corresponding to the automobile insurance policy information, and marking the vehicle driving trajectory data as normal or abnormal;

[0012] extracting the claim settlement features of the motor vehicle insurance policy information, and extracting the trajectory features according to the normal or abnormal labels of the vehicle driving trajectory data;

[0013] An auto insurance anomaly detection model is constructed using the isolation forest algorithm according to the claim characteristics and the trajectory characteristics.

[0014] Optionally, the step of constructing a car insurance anomaly detection model using an isolation forest algorithm according to the claim settlement feature and the trajectory feature includes:

[0015] Using the claim settlement feature and the trajectory feature as a data point set of an isolation forest model, and initializing the number of isolated trees to a preset value;

[0016] For each isolated tree, randomly sampling a data point from the data point set for constructing the isolated tree;

[0017] For each data point, calculate its isolation score on each isolated tree;

[0018] The average score of all data points is calculated based on the isolation score of each data point. If the difference between the isolation score of a data point and the average score is within a preset difference anomaly threshold, the data point is marked as abnormal, and the auto insurance anomaly detection model is constructed.

[0019] Optionally, before the step of marking a data point as abnormal if the difference between the isolation score of a data point and the average score is within a preset difference abnormality threshold, the step further includes:

[0020] Set the preset difference abnormality threshold to a range of 0-15, or set the preset difference abnormality threshold to a range of greater than 100.

[0021] Optionally, before the step of building the vehicle insurance anomaly detection model, the step further includes:

[0022] The isolation forest model is trained cyclically until the number of training times for each data point in the data point set reaches a preset number of times.

[0023] Optionally, after the step of constructing a car insurance anomaly detection model using an isolation forest algorithm according to the claim settlement features and the trajectory features, the method further includes:

[0024] A test sample set is obtained, wherein the ratio of positive and negative samples in the test sample set is determined according to the motor vehicle insurance policy information, and the performance of the motor vehicle insurance anomaly detection model is evaluated using the test sample set.

[0025] Optionally, the step of obtaining the automobile insurance policy information includes:

[0026] The motor vehicle insurance policy data of a preset year is obtained, the motor vehicle insurance policy data is cleaned, duplicate data and data with missing information in the motor vehicle insurance policy data are deleted, and the motor vehicle insurance policy information is obtained.

[0027] Optionally, the step of marking the vehicle driving trajectory data as normal or abnormal includes:

[0028] Determine the insurance policy information corresponding to the vehicle driving trajectory data, and judge whether the vehicle driving trajectory corresponding to the insurance policy information is normal when the insurance policy information occurs;

[0029] If so, marking the vehicle driving trajectory data as normal;

[0030] If not, the vehicle driving trajectory data is marked as abnormal.

[0031] Furthermore, in order to achieve the above-mentioned purpose, the present invention also provides a device for constructing a vehicle insurance anomaly detection model, and the device for constructing a vehicle insurance anomaly detection model comprises:

[0032] An information acquisition module, which acquires the automobile insurance policy information, acquires the vehicle driving trajectory data corresponding to the automobile insurance policy information, and marks the vehicle driving trajectory data as normal or abnormal;

[0033] A feature extraction module, extracting claim settlement features of the motor vehicle insurance policy information, and extracting trajectory features according to normal or abnormal labels of the vehicle driving trajectory data;

[0034] The model building module uses the isolation forest algorithm to build a car insurance anomaly detection model according to the claim characteristics and the trajectory characteristics.

[0035] Furthermore, to achieve the above-mentioned purpose, the present invention also provides a device for constructing a vehicle insurance anomaly detection model, the device for constructing a vehicle insurance anomaly detection model comprising a memory, a processor, and a construction program for the vehicle insurance anomaly detection model stored in the memory and executable on the processor, and the construction program for the vehicle insurance anomaly detection model, when executed by the processor, implements the steps of the method for constructing a vehicle insurance anomaly detection model as described above.

[0036] Furthermore, to achieve the above-mentioned purpose, the present invention also provides a storage medium, on which is stored a program for constructing a vehicle insurance anomaly detection model. When the program for constructing a vehicle insurance anomaly detection model is executed by a processor, the steps of the method for constructing a vehicle insurance anomaly detection model as described above are implemented.

[0037] The present invention relates to a method for constructing a motor vehicle insurance anomaly detection model, by obtaining motor vehicle insurance policy information and simultaneously obtaining vehicle driving trajectory data corresponding to the motor vehicle insurance policy information, marking the vehicle driving trajectory data as normal or abnormal. The claim settlement features of the motor vehicle insurance policy information are extracted, the trajectory features are extracted according to the normal or abnormal labels of the vehicle driving trajectory data, and the motor vehicle insurance anomaly detection model is constructed using an isolation forest algorithm according to the claim settlement features and the trajectory features. The present invention constructs multiple randomized binary trees through an isolation forest algorithm, which can effectively identify abnormal data points, significantly improves the detection accuracy of abnormal data in motor vehicle insurance claims, and can complete the processing and analysis of large-scale data in a relatively short time, which is suitable for the detection needs of massive claims data in actual business. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 A schematic diagram of the structure of the device hardware operating environment involved in the embodiment of the device for building the vehicle insurance anomaly detection model of the present invention;

[0039] Figure 2 It is a flowchart of a first embodiment of a method for constructing a vehicle insurance anomaly detection model according to the present invention;

[0040] Figure 3 This is a schematic diagram of the functional modules of a preferred embodiment of the device for constructing a vehicle insurance anomaly detection model of the present invention.

[0041] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0042] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0043] It should be noted that with the increase in the number of cars, the auto insurance business has developed rapidly. However, there are a lot of frauds in the auto insurance claims process, such as false reporting of accidents, forged evidence, exaggeration of losses, etc. These frauds not only increase the financial burden of insurance companies, but also disrupt the normal order of the insurance market. Therefore, how to effectively identify and prevent auto insurance fraud has become an urgent problem that the insurance industry needs to solve.

[0044] Traditional auto insurance fraud detection methods mainly rely on manual review and simple rule-based judgment. The shortcomings of these methods are:

[0045] Inefficiency: Manual review requires a lot of manpower and time, and it is difficult to cope with a large number of claims cases.

[0046] Limited accuracy: Rule-based detection methods can only capture known fraud patterns and are often helpless against new or mutated fraud behaviors.

[0047] Strong subjectivity: Manual review is subject to subjective judgment and is easily affected by the reviewer’s experience and emotions.

[0048] In recent years, with the development of big data and artificial intelligence technology, data-driven fraud detection methods have gradually gained attention. Through machine learning algorithms, patterns of fraudulent behavior can be automatically learned and discovered from massive historical data, thereby improving the efficiency and accuracy of detection. As an effective anomaly detection algorithm, the isolation forest algorithm has been widely used in fraud detection due to its high computational efficiency and superior detection performance.

[0049] The Isolation Forest Algorithm is an unsupervised learning method based on a tree model. It measures the isolation of data points by constructing multiple randomized binary trees. The core idea of ​​the Isolation Forest Algorithm is that outliers are usually isolated at a lower level in the tree structure, that is, the path is shorter. Therefore, the path length of the data point in the tree can be calculated to determine whether it is an outlier.

[0050] Based on the above-mentioned defects, the present invention provides a device for constructing a vehicle insurance anomaly detection model, referring to Figure 1 , Figure 1 This is a schematic diagram of the structure of the device hardware operating environment involved in the embodiment of the device for building a car insurance anomaly detection model of the present invention.

[0051] like Figure 1 As shown, the construction device of the vehicle insurance anomaly detection model may include: a processor 1001, such as a CPU, a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. Among them, the communication bus 1002 is used to realize the connection and communication between these components. The user interface 1003 may include a display screen (Display), an input unit such as a keyboard (Keyboard), and the optional user interface 1003 may also include a standard wired interface and a wireless interface. The network interface 1004 may optionally include a standard wired interface and a wireless interface (such as a WI-FI interface). The memory 1005 may be a high-speed RAM memory, or a stable memory (non-volatile memory), such as a disk memory. The memory 1005 may also be a storage device independent of the aforementioned processor 1001.

[0052] Those skilled in the art will understand that Figure 1The hardware structure of the device for building a vehicle insurance anomaly detection model shown in the figure does not constitute a limitation on the device for building a vehicle insurance anomaly detection model, and may include more or fewer components than shown in the figure, or a combination of certain components, or a different arrangement of components.

[0053] like Figure 1 As shown, the memory 1005 as a storage medium may include an operating system, a network communication module, a user interface module, and a program for constructing a vehicle insurance anomaly detection model. The operating system is a program for managing and controlling the construction equipment and software resources of the vehicle insurance anomaly detection model, and supports the operation of the network communication module, the user interface module, the program for constructing the vehicle insurance anomaly detection model, and other programs or software; the network communication module is used to manage and control the network interface 1004; and the user interface module is used to manage and control the user interface 1003.

[0054] exist Figure 1 In the hardware structure of the construction device of the vehicle insurance anomaly detection model shown, the network interface 1004 is mainly used to connect to the backend server and communicate data with the backend server; the user interface 1003 is mainly used to connect to the client and communicate data with the client; the processor 1001 can call the construction program of the vehicle insurance anomaly detection model stored in the memory 1005 and perform the following operations:

[0055] Obtaining automobile insurance policy information, obtaining vehicle driving trajectory data corresponding to the automobile insurance policy information, and marking the vehicle driving trajectory data as normal or abnormal;

[0056] extracting the claim settlement features of the motor vehicle insurance policy information, and extracting the trajectory features according to the normal or abnormal labels of the vehicle driving trajectory data;

[0057] An auto insurance anomaly detection model is constructed using the isolation forest algorithm according to the claim characteristics and the trajectory characteristics.

[0058] Furthermore, the step of constructing a car insurance anomaly detection model using an isolation forest algorithm according to the claim settlement features and the trajectory features includes:

[0059] Using the claim settlement feature and the trajectory feature as a data point set of an isolation forest model, and initializing the number of isolated trees to a preset value;

[0060] For each isolated tree, randomly sampling a data point from the data point set for constructing the isolated tree;

[0061] For each data point, calculate its isolation score on each isolated tree;

[0062] The average score of all data points is calculated based on the isolation score of each data point. If the difference between the isolation score of a data point and the average score is within a preset difference anomaly threshold, the data point is marked as abnormal, and the auto insurance anomaly detection model is constructed.

[0063] Furthermore, if the difference between the isolation score of a data point and the average score is within a preset difference abnormality threshold, before the step of marking the data point as abnormal, the step further includes:

[0064] Set the preset difference abnormality threshold to a range of 0-15, or set the preset difference abnormality threshold to a range of greater than 100.

[0065] Furthermore, before the step of constructing the vehicle insurance anomaly detection model, it also includes:

[0066] The isolation forest model is trained cyclically until the number of training times for each data point in the data point set reaches a preset number of times.

[0067] Furthermore, after the step of constructing a car insurance anomaly detection model using an isolation forest algorithm according to the claim settlement features and the trajectory features, the method further includes:

[0068] A test sample set is obtained, wherein the ratio of positive and negative samples in the test sample set is determined according to the motor vehicle insurance policy information, and the performance of the motor vehicle insurance anomaly detection model is evaluated using the test sample set.

[0069] Furthermore, the step of obtaining the automobile insurance policy information includes:

[0070] The motor vehicle insurance policy data of a preset year is obtained, the motor vehicle insurance policy data is cleaned, duplicate data and data with missing information in the motor vehicle insurance policy data are deleted, and the motor vehicle insurance policy information is obtained.

[0071] Furthermore, the step of marking the vehicle driving trajectory data as normal or abnormal includes:

[0072] Determine the insurance policy information corresponding to the vehicle driving trajectory data, and judge whether the vehicle driving trajectory corresponding to the insurance policy information is normal when the insurance policy information occurs;

[0073] If so, marking the vehicle driving trajectory data as normal;

[0074] If not, the vehicle driving trajectory data is marked as abnormal.

[0075] The specific implementation of the device for constructing the abnormal vehicle insurance detection model of the present invention is basically the same as the various embodiments of the method for constructing the abnormal vehicle insurance detection model described below, and will not be repeated here.

[0076] The present invention also provides a method for constructing a vehicle insurance anomaly detection model based on the above-mentioned vehicle insurance anomaly detection model construction device.

[0077] Reference Figure 2 , Figure 2 This is a flow chart of the first embodiment of the method for constructing a vehicle insurance anomaly detection model of the present invention.

[0078] An embodiment of the present invention provides an embodiment of a method for constructing a vehicle insurance anomaly detection model. It should be noted that although a logical order is shown in the flow chart, in some cases, the steps shown or described may be performed in an order different from that shown here.

[0079] In various embodiments of the method for constructing a vehicle risk anomaly detection model, the execution subject is a device for constructing a vehicle risk anomaly detection model.

[0080] The method for constructing a vehicle insurance anomaly detection model is applied to a device for constructing a vehicle insurance anomaly detection model. The method for constructing a vehicle insurance anomaly detection model includes:

[0081] Step S10, obtaining automobile insurance policy information, obtaining vehicle driving trajectory data corresponding to the automobile insurance policy information, and marking the vehicle driving trajectory data as normal or abnormal;

[0082] With the increase in the number of cars, the auto insurance business has developed rapidly. However, there are a lot of frauds in the process of auto insurance claims, such as false reporting of accidents, forged evidence, exaggeration of losses, etc. These frauds not only increase the financial burden of insurance companies, but also disrupt the normal order of the insurance market. Therefore, how to effectively identify and prevent auto insurance fraud has become an urgent problem that the insurance industry needs to solve.

[0083] Traditional auto insurance fraud detection methods mainly rely on manual review and simple rule-based judgment. The shortcomings of these methods are:

[0084] Inefficiency: Manual review requires a lot of manpower and time, and it is difficult to cope with a large number of claims cases.

[0085] Limited accuracy: Rule-based detection methods can only capture known fraud patterns and are often helpless against new or mutated fraud behaviors.

[0086] Strong subjectivity: Manual review is subject to subjective judgment and is easily affected by the reviewer’s experience and emotions.

[0087] In recent years, with the development of big data and artificial intelligence technologies, data-driven fraud detection methods have gradually gained attention. Through machine learning algorithms, patterns of fraud behavior can be automatically learned and discovered from a vast amount of historical data, thereby improving the efficiency and accuracy of detection. As an effective anomaly detection algorithm, the Isolation Forest algorithm has been widely applied in fraud detection due to its high computational efficiency and excellent detection performance.

[0088] The Isolation Forest algorithm is an unsupervised learning method based on tree models, which measures the isolation degree of data points by constructing multiple randomized binary trees. The core idea of the Isolation Forest algorithm is that outliers are usually isolated at lower levels in the tree structure, that is, the paths are shorter. Therefore, whether a data point is an outlier can be judged by calculating the path length of the data point in the tree.

[0089] In this embodiment, data preparation is carried out first.

[0090] Collect auto insurance policies (claims information) for several years, including vehicle information, claim amount, accident date and location, etc. As training samples, those with a claim settlement conclusion of rejection are set as positive samples, and those with a claim settlement are set as negative samples.

[0091] The specific implementation method can be: export auto insurance policy data and claim data for several years from the insurance company database. For example, collect data for certain years, and the specific data includes but is not limited to vehicle information (such as vehicle identification number, license plate number, vehicle make and model, registration date, etc.) and claim information (such as vehicle identification number, claim amount, accident date and location, claim settlement conclusion, etc.). The vehicle identification number can be used as the key field to match the policy data with the claim data to ensure that each piece of data corresponds to the correct vehicle and claim information.

[0092] Example:

[0093] The collected data examples are as follows:

[0094] Vehicle identification number: LJWFF3A1XK1234567

[0095] License plate number: Shanghai A12345

[0096] Vehicle make and model: Toyota Camry

[0097] Registration date: 2018-05-20

[0098] Claim amount: 5000 yuan

[0099] Accident date: 2020-06-15

[0100] Accident location: Xuhui District, Shanghai

[0101] Claim settlement conclusion: Rejection.

[0102] Obtain the vehicle driving trajectory data corresponding to the auto insurance policy information, and determine whether it conforms to normal driving behavior based on the trajectory data at the time of the accident, and mark it as normal or abnormal.

[0103] Step S20, extracting the claim settlement features of the motor vehicle insurance policy information, and extracting the trajectory features according to the normal or abnormal labels of the vehicle driving trajectory data;

[0104] In this embodiment, the insured person information (such as age, gender, occupation), vehicle information (such as brand, model, age), historical accident records, etc. are extracted from the insurance policy data.

[0105] Extract claim amount, accident frequency, claim conclusion, etc. from claim data.

[0106] Example:

[0107] The extracted claim features are as follows:

[0108] Age of the insured: 45 years old

[0109] Gender: Male

[0110] Occupation: Company employee

[0111] Vehicle brand: Toyota Camry

[0112] Use life: 4 years

[0113] Historical accident records: 2 times

[0114] Claim amount: 5,000 yuan

[0115] Accident frequency: 0.5 times per year

[0116] Claim settlement conclusion: denied.

[0117] Based on the vehicle driving trajectory data, features such as driving speed, driving path, and stay time are extracted.

[0118] According to the normal or abnormal labels of the vehicle driving trajectory data, features such as driving speed, driving path, and dwell time are extracted.

[0119] Example: The extracted trajectory features are as follows:

[0120] Driving speed: 100 km / h

[0121] Driving path: Abnormal

[0122] Stay time: No abnormality

[0123] Normal or abnormal label: abnormal.

[0124] Step S30, constructing a car insurance anomaly detection model using an isolation forest algorithm according to the claim characteristics and the trajectory characteristics.

[0125] In this embodiment, the extracted claim features and trajectory features may be standardized to ensure that the feature values ​​are within a uniform range. For example, all numerical features may be standardized by z-score.

[0126] Use the Isolation Forest algorithm to build a model, and use the claim characteristics and trajectory characteristics as input data for training. Example: Use the IsolationForest class in the Python library to build an isolation forest model.

[0127] The specific steps may include: using the claim characteristics and the trajectory characteristics as a data point set of the isolation forest model, and initializing the number of isolated trees to a preset value.

[0128] Initialize isolated trees: Set the number of isolated trees to 10. Example: Set the parameter n_estimators = 10 in the code to initialize 10 isolated trees.

[0129] Build isolated trees: Randomly sample data points from the training dataset and build each isolated tree. Example: Use the Isolation Forest algorithm to automate the process of building each isolated tree.

[0130] Calculate the isolation score: For each data point, calculate its isolation score on each isolated tree. Example: Using the decision_function method of the Isolation Forest algorithm, calculate the isolation score for each data point.

[0131] Mark abnormal data points: Mark abnormal data points based on the difference between the isolation score and the average score of each data point. Example: Mark data points whose difference between the isolation score and the average score is greater than 15 or less than 0 as abnormal.

[0132] Set a difference anomaly threshold: Set a reasonable difference anomaly threshold based on business requirements and data distribution. Example: Set the difference anomaly threshold to 0-15, or to a value greater than 100.

[0133] Cyclic training: Cyclic training of the isolation forest model is performed until the number of training times for each data point reaches a preset number (such as 300 rounds). Example: During the training process, the number of training times for each data point is recorded to ensure that each data point participates in at least 300 rounds of training.

[0134] The present invention relates to a method for constructing a motor vehicle insurance anomaly detection model, by obtaining motor vehicle insurance policy information and simultaneously obtaining vehicle driving trajectory data corresponding to the motor vehicle insurance policy information, marking the vehicle driving trajectory data as normal or abnormal. The claim settlement features of the motor vehicle insurance policy information are extracted, the trajectory features are extracted according to the normal or abnormal labels of the vehicle driving trajectory data, and the motor vehicle insurance anomaly detection model is constructed using an isolation forest algorithm according to the claim settlement features and the trajectory features. The present invention constructs multiple randomized binary trees through an isolation forest algorithm, which can effectively identify abnormal data points, significantly improves the detection accuracy of abnormal data in motor vehicle insurance claims, and can complete the processing and analysis of large-scale data in a relatively short time, which is suitable for the detection needs of massive claims data in actual business.

[0135] Furthermore, a second embodiment of the method for constructing a vehicle insurance anomaly detection model of the present invention is proposed, wherein the above step S30 includes:

[0136] Step a1, using the claim settlement feature and the trajectory feature as a data point set of the isolation forest model, and initializing the number of isolated trees to a preset value;

[0137] Step a2: for each isolated tree, randomly sampling a data point from the data point set for constructing the isolated tree;

[0138] Step a3, for each data point, calculate its isolation score on each isolated tree;

[0139] Step a4, calculate the average score of all data points based on the isolation score of each data point. If the difference between the isolation score of a data point and the average score is within a preset difference anomaly threshold, mark the data point as abnormal and build the auto insurance anomaly detection model.

[0140] In this embodiment, the number of isolated trees needs to be set, for example, 10, that is, 10 isolated trees are initialized. The number of isolated trees is an important parameter of the algorithm, which affects the performance and computational efficiency of the model. Generally speaking, more isolated trees can improve the stability and accuracy of the model, but will also increase the computational overhead.

[0141] Randomly sample data points from the training data set for the construction of each isolated tree. The Isolation Forest algorithm measures the isolation of data points by constructing multiple randomized binary trees. Therefore, when constructing each tree, data points need to be randomly sampled to ensure the randomness and diversity of the tree.

[0142] In order to build the Isolation Forest Model, it is necessary to combine the claim features and trajectory features into a comprehensive set of data points. These features contain various information about the auto insurance policy and data related to the vehicle's driving trajectory. The specific steps can be as follows:

[0143] Feature extraction: Extract claim features from auto insurance policy information, including policyholder information (such as age, gender, occupation), vehicle information (such as brand, model, age), historical accident records, claim amount, accident frequency, and claim conclusions. Extract trajectory features from vehicle driving trajectory data, including driving speed, driving path, dwell time, and labels indicating whether the driving trajectory is normal.

[0144] Feature merging: Merge the extracted claims features and trajectory features to form a comprehensive set of data points. Each data point contains all the extracted claims features and trajectory features as input data for the isolation forest model.

[0145] Standardization: In order to ensure the effectiveness of model training, all features need to be standardized so that they are in the same numerical range.

[0146] Example: Claim characteristics: insured age, gender, occupation, vehicle brand, age, historical accident records, claim amount, accident frequency, claim conclusion, etc. Trajectory characteristics: driving speed, driving path, dwell time, normal or abnormal label.

[0147] Example of a merged set of data points:

[0148] Data point 1: Age 45, gender male, occupation company employee, brand Toyota Camry, usage period 4 years, historical accident records 2 times, claim amount 5,000 yuan, accident frequency 0.5 times / year, claim settlement rejected, driving speed 50 km / h, driving route normal, no abnormal stop time, normal label.

[0149] Data point 2: age 32, gender female, occupation teacher, brand Honda Accord, usage period 2 years, historical accident record 0 times, claim amount 3,000 yuan, accident frequency 0.2 times / year, claim settlement conclusion claim, driving speed 40 km / h, abnormal driving path, abnormal stop time, abnormal label.

[0150] Calculate the isolation score: For each data point, calculate its isolation score on each isolated tree. The isolation score indicates the degree to which the data point is isolated. The smaller the score, the more likely the data point is an outlier. The specific implementation method can be:

[0151] The calculation formula for the isolation score is:

[0152]

[0153] Among them, EPT(x) is the expected path length (Expected Path Length) of data point x, l(x) is the number of leaf node layers of data point x in the decision tree, and c is the average path length of the decision tree.

[0154] The expected path length formula is:

[0155] EPT(x)=c(n)+l(x)

[0156] Among them, l(x) is the number of layers (path length) of leaf nodes of data point x in the decision tree, and c(n) is a constant determined by the size of the data set n. where H(i) is the i-th harmonic number, ≈log(i)+0.5772156649 (Euler-Mascheroni constant).

[0157] EPT(x) represents the expected path length of data point x in the isolation tree. The lower the isolation score, the easier it is for data point x to be isolated, indicating that it is more likely to be an outlier. The average path length c of the decision tree is a constant determined by the size of the data set n and is used to normalize the path length.

[0158] Calculate the average isolation score of all data points: Calculate the average of the isolation scores of all data points, recorded as Mean Isolation Score.

[0159] Calculate the difference: For each data point, calculate the difference between its isolation score and the average isolation score, denoted as Δ:

[0160] Δ=∣Isolation Score(x)-Mean Isolation Score∣

[0161] Set difference threshold: Set the difference anomaly threshold range of the isolation score, for example, the difference anomaly threshold is 0-15, or greater than 100, and mark abnormal data points according to the difference threshold.

[0162] Cyclic training: The isolation forest model is trained cyclically until the number of training times for each data point reaches a preset number (such as 300 rounds), thereby building a car insurance anomaly detection model.

[0163] In this embodiment, the algorithm shows strong robustness and stability in processing data noise and unbalanced data by constructing multiple isolated trees and calculating isolation scores. Even in the case of unbalanced data volume, it can maintain high detection accuracy and stability. Through standardization processing, cyclic training and other steps, the performance and detection effect of the model are further improved.

[0164] Furthermore, a third embodiment of the method for constructing a vehicle insurance anomaly detection model of the present invention is proposed. In the above step a4, before the step of marking a data point as abnormal if the difference between the isolation score of a data point and the average score is within a preset difference abnormality threshold, the step further includes:

[0165] Step b1, setting the preset difference abnormality threshold to a range of 0-15, or setting the preset difference abnormality threshold to a range of greater than 100.

[0166] In this embodiment, the main function of setting the difference threshold is to identify abnormal points in the data by the difference between the isolation score and the average score. The difference threshold can help distinguish normal data points from potential abnormal data points, thereby improving the accuracy and effectiveness of detection. By setting a reasonable difference threshold, the false alarm rate (i.e., the situation where normal points are mistakenly judged as abnormal points) can be reduced and the accuracy of the model can be improved. An appropriate threshold can ensure that only those truly abnormal points are marked as abnormal, avoiding too many false alarms. The preset difference threshold can improve the stability and robustness of the model to a certain extent, especially when processing data sets of different types and sizes, and can provide consistent anomaly detection results.

[0167] The methods for setting the difference anomaly threshold include: fixed threshold setting, dynamic threshold setting, percentile method, etc.

[0168] Fixed threshold setting: Preset a fixed difference threshold range, such as 0-15 and greater than 100. This method is simple and intuitive, and is suitable for situations where data characteristics and distribution are relatively stable.

[0169] Dynamic threshold setting: Dynamically adjust the threshold according to the statistical characteristics of the data distribution. For example, you can use statistics such as the standard deviation and interquartile range of the data to determine the adaptive difference threshold. You can use the standard deviation method: set the threshold to the mean plus or minus a number of standard deviations. For example, set the threshold to the mean plus or minus 2 standard deviations, that is, μ±2σ. Or use the quartile method: set the threshold to the data point outside the quartile range. For example, set the threshold to [Q1-1.5×IQR,Q3+1.5×IQR], where IQR=Q3-Q1.

[0170] Among them, Q1 is the first quartile of the data set, also known as the lower quartile. It means that 25% of the data points in the data set are below Q1, that is, the data points in the first 25% after sorting. Q1 is used to represent the lower bound of the data, help identify lower values ​​in the data set, and provide a reference for outlier detection.

[0171] Q3 is the third quartile of the data set, also known as the upper quartile. It means that 75% of the data points in the data set are below Q3, that is, the data points in the first 75% after sorting. Q3 is used to represent the upper bound of the data, help identify higher values ​​in the data set, and provide a reference for outlier detection.

[0172] IQR is the interquartile range, which represents the span of the middle 50% of the data set, specifically the difference between Q3 and Q1. IQR = Q3-Q1, which is used to measure the dispersion and variability of data. A larger IQR means that the data is more widely distributed and has greater variability; a smaller IQR means that the data is more concentrated and has less variability.

[0173] 1.5×IQR is an empirical coefficient used to expand the range of the interquartile range to identify outliers. This multiple is used to relax the range of Q1 and Q3 to identify outliers that are far away from the middle 50% of the data. 1.5×IQR is widely used in outlier detection and can effectively balance the sensitivity of outlier detection, identifying outliers without over-excluding normal values.

[0174] [Q1-1.5×IQR,Q3+1.5×IQR] represents the outlier detection interval:

[0175] Lower threshold (Q1-1.5×IQR): Data points below this value are usually considered lower bound outliers. The lower threshold is used to detect and exclude outliers in a data set that are significantly lower than most data points.

[0176] Upper threshold (Q3 + 1.5 × IQR): Data points above this value are usually considered to be upper bound outliers. The upper threshold is used to detect and exclude outliers in the data set that are significantly higher than most data points.

[0177] Example description:

[0178] Suppose there is a set of data: 3, 7, 8, 12, 13, 14, 18, 21, 24, 100.

[0179] Q1 (first quartile): the first 25% of the data after sorting (the average of the 3rd and 4th numbers): 7+8 / 2=7.5;

[0180] Q3 (third quartile): the first 75% of the position data after sorting (the average of the 7th and 8th numbers): 18+21 / 2=19.5;

[0181] IQR (interquartile range): IQR = Q3-Q1 = 19.5-7.5 = 12;

[0182] Lower threshold (Q1-1.5×IQR): 7.5-1.5×12=-10.5;

[0183] Upper threshold (Q3+1.5×IQR): 19.5+1.5×12=37.5;

[0184] Based on the calculation, any value less than -10.5 or greater than 37.5 will be considered an outlier. The value 100 in the dataset is greater than the upper threshold of 37.5, so it is considered an outlier.

[0185] Percentile method: The threshold is determined based on the percentile of the data point. For example, anomalies can be defined as data points with scores in the top 5% or bottom 5%, the percentile of the scores is calculated, and the scores in the top 5% and bottom 5% are set as the anomaly thresholds.

[0186] Through the above steps, this embodiment can reasonably set the difference threshold and effectively mark the abnormal points in the motor insurance claim data, thereby improving the accuracy and efficiency of anomaly detection.

[0187] Furthermore, a fourth embodiment of the method for constructing a vehicle insurance anomaly detection model of the present invention is proposed. In the above step a4, before the step of constructing the vehicle insurance anomaly detection model, the method further includes:

[0188] Step c1, cyclically training the isolation forest model until the number of training times for each data point in the data point set reaches a preset number of times.

[0189] In this embodiment, loop training refers to repeatedly training the training data set during the construction of the isolation forest model until the number of training times for each data point reaches a preset threshold. Loop training plays the following important roles in the auto insurance anomaly detection model:

[0190] Improve model stability: Through multiple training, the model can learn the patterns and features in the data more fully, improving the stability and robustness of the model. Single training may be affected by data noise and randomness, while cyclic training can reduce these effects.

[0191] Enhance the generalization ability of the model: Cyclic training enables the model to be trained multiple times on different data subsets, enhancing the generalization ability of the model. This ensures that the model can still maintain good detection performance when facing new and unseen data.

[0192] Make full use of training data:

[0193] Cyclic training can ensure that each data point participates in the model training process multiple times, making full use of the training data and improving the model's learning effect on data features. Especially when the amount of data is large, cyclic training can ensure that all data points are fully trained.

[0194] Reduce the risk of overfitting: Through multiple training, the risk of overfitting the model to specific training data can be reduced. Multiple training allows the model to learn the common features in the data instead of remembering the noise and anomalies of specific data points.

[0195] Specific implementation methods may include:

[0196] Initialization training: Initially train the isolation forest model and set the number of isolated trees to a preset value (for example, 10).

[0197] Record the number of training times for each data point: Initialize a counter to record the number of training times for each data point, with an initial value of 0.

[0198] Cyclic training: The isolation forest model is trained cyclically until the number of training times for each data point reaches a preset number (for example, 300 rounds). After each training, the number of training times for each data point is updated.

[0199] For example, randomly sample data points from the training data set for the construction of each isolated tree, and use the isolation forest algorithm to automatically complete the construction process of each isolated tree. For each data point, calculate its isolation score on each isolated tree. After each training, increase the number of training times for each data point to determine whether the number of training times for each data point reaches the preset number. If not, continue the training cycle.

[0200] Through cyclic training, this embodiment can significantly improve the performance and stability of the car insurance anomaly detection model, enhance the model's ability to detect abnormal data points, and thus more effectively identify and prevent abnormal behaviors in the car insurance claims process.

[0201] Furthermore, a fifth embodiment of the method for constructing a vehicle insurance anomaly detection model of the present invention is proposed, and after the above step S30, the method further includes:

[0202] Step d1, obtaining a test sample set, wherein the ratio of positive and negative samples in the test sample set is determined according to the motor vehicle insurance policy information, and the test sample set is used to evaluate the performance of the motor vehicle insurance anomaly detection model.

[0203] In this embodiment, a test sample set is taken and the performance of the auto insurance anomaly detection model is evaluated. The test samples and training samples are usually from the same data set or similar data sources. In the process of building the auto insurance anomaly detection model, both the training samples and the test samples are derived from auto insurance policy information and claims data, including vehicle information, accident date and location, claim amount, driving trajectory, etc. The specific steps may include:

[0204] Determine the test sample set: Select test samples from the data set based on the auto insurance policy information. The test samples should include positive samples (abnormal cases) and negative samples (normal cases) to fully evaluate the model performance.

[0205] Set the ratio of positive and negative samples: Determine the ratio of positive and negative samples based on the historical data in the auto insurance policy information. For example, a positive sample can be a case where the claim conclusion is rejected, and a negative sample can be a case where the claim conclusion is approved. Assume that the ratio of positive and negative samples is 1:10, that is, for every 1 positive sample, there are 10 negative samples.

[0206] Example:

[0207] Suppose there is a car insurance dataset containing policy and claim information from 2019 to 2023. When building and evaluating the model, you can follow the following steps:

[0208] Data preparation:

[0209] Training samples: Extract training samples from data from 2019 to 2022. Training samples include cases where the claim conclusion is rejected (positive samples) and cases where the claim is settled (negative samples). Example: Select 90,000 positive samples and 4,000,000 negative samples from data from 2019 to 2022.

[0210] Test samples: Extract test samples from the data of 2023. The positive and negative sample ratio of the test samples is set to 1:10. Example: Select 1000 positive samples and 10000 negative samples from the data of 2023.

[0211] After obtaining the test sample set, the auto insurance anomaly detection model is evaluated based on the test sample set. The specific steps may include:

[0212] Determine the evaluation indicators: Use the test sample set to evaluate the performance of the auto insurance anomaly detection model. Common evaluation indicators include accuracy, recall, precision, and F1 score.

[0213] Predict test samples: Use the constructed auto insurance anomaly detection model to predict the test sample set and obtain the prediction result (normal or abnormal) for each sample.

[0214] Calculate evaluation indicators: Calculate the evaluation indicators of the model based on the prediction results and actual labels.

[0215] Example: Assume that the evaluation results of the model on the test sample are: accuracy 95%, recall 90%, precision 80%, and F1 score 85%.

[0216] In addition, the Aether platform is used for the above model training and production deployment. The production caller is the auto insurance claims system. The model interface needs to be called in all aspects of claims: reporting, investigation, and damage assessment. The factor name is reporting risk model-national, national institution configuration, and the rule scenarios are: reporting and adjustment rules, damage assessment and adjustment rules, investigation and adjustment rules, whole case investigation and adjustment, personal injury investigation and adjustment rules, and vehicle damage assessment and adjustment. The rule group is set with the following first: AI_Placement site fraud model, AI_Repeated accident model, national drunk driving fraud risk model, placement site fraud model, fake license plate placement CK, general risk model (AI), night risk model CK, daytime risk model CK, commercial exclusive risk model, commercial truck anti-fraud model, [Drunk driving] drunk driving fraud risk model, national drunk driving, personal injury investigation risk model. The factor value is set to: 490 points.

[0217] The accuracy rate has been greatly improved, from 21% to 29%. The model has a data refresh effect in October 2023, and the trigger volume can be increased to 854, the number of dispatched cases is 664, the number of rejected claims is 264, the rejected claim amount (10,000) is 1,183, the dispatch rate is 57.48%, the investigation success rate is 47.56%, and the rejection rate is 27.34%.

[0218] In this embodiment, the test samples are used to verify the performance of the trained model on unseen data to ensure the effectiveness and reliability of the model in practical applications.

[0219] Furthermore, a sixth embodiment of the method for constructing a vehicle insurance anomaly detection model of the present invention is proposed. In the above step S10, the step of obtaining vehicle insurance policy information includes:

[0220] Step e1, obtaining the automobile insurance policy data of a preset year, performing data cleaning on the automobile insurance policy data, deleting duplicate data and data with missing information in the automobile insurance policy data, and obtaining the automobile insurance policy information.

[0221] In this embodiment, the step of obtaining the automobile insurance policy information may include:

[0222] Get the car insurance policy data of the preset year: Export the car insurance policy data of the preset year from the insurance company's database. The preset year can be any year or several years in the past, and the specific year is determined according to actual needs. For example, you can select data from 2019 to 2023.

[0223] Data cleaning: Perform data cleaning on the exported auto insurance policy data to ensure the integrity and accuracy of the data. The specific steps for data cleaning are as follows:

[0224] De-duplicate data: Check for duplicate records in the auto insurance policy data and delete these duplicate records. Duplicate data may be caused by data entry errors or system failures. De-duplicate data can avoid data redundancy and errors in model training.

[0225] Processing missing data: Processing missing data in the auto insurance policy data. Missing data may affect the training effect of the model, so appropriate methods need to be adopted to process it. Common methods for processing missing data include: Deleting missing values: For data with a small missing ratio, you can directly delete the records containing missing values. Filling missing values: For data with a large missing ratio, you can use the mean filling method, interpolation method or KNN filling method to fill missing values.

[0226] Standardized data format: Ensure that all fields in the auto insurance policy data are in the same format, such as date format, amount unit, etc. Standardized data format helps improve the accuracy of data processing and analysis.

[0227] Check for outliers: Check for outliers in the auto insurance policy data and handle them accordingly. Outliers may be caused by data entry errors or other reasons. Outliers can be detected and handled through methods such as IQR method and Z score.

[0228] After data cleaning, complete and accurate auto insurance policy information is obtained. The cleaned data can be used for subsequent feature extraction and model training.

[0229] This embodiment obtains complete and accurate auto insurance policy information through data cleaning, ensures data quality, and provides a reliable data foundation for subsequent feature extraction and model training.

[0230] Furthermore, a seventh embodiment of the method for constructing a vehicle insurance anomaly detection model of the present invention is proposed. In the above step S10, the step of marking the vehicle driving trajectory data as normal or abnormal includes:

[0231] Step f1, determining the insurance policy information corresponding to the vehicle driving trajectory data, and judging whether the vehicle driving trajectory corresponding to the insurance policy information is normal when the insurance policy information is in danger;

[0232] Step f2: if yes, marking the vehicle driving trajectory data as normal;

[0233] Step f3: if not, mark the vehicle driving trajectory data as abnormal.

[0234] In this embodiment, the specific steps may include:

[0235] Determine the insurance policy information corresponding to the vehicle driving trajectory data: Match the vehicle driving trajectory data with the corresponding vehicle insurance policy information according to the vehicle identification number or other unique identifier. Ensure that each trajectory data has a corresponding policy record.

[0236] Judge whether the vehicle driving trajectory data is normal: Analyze the characteristics of the vehicle driving trajectory data and judge whether the corresponding vehicle driving trajectory at the time of accident (when the accident occurs) conforms to the normal driving behavior pattern. The judgment criteria can include but are not limited to the following aspects: Driving speed: Whether the vehicle is driving within a reasonable speed range. Driving route: Whether the vehicle is driving along the expected route, and whether there are abnormal detours or deviations from the route.停留时间: Whether the vehicle stays at an abnormal time period or location for too long. Driving behavior: Whether the vehicle has abnormal driving behaviors such as sudden braking or sharp turning at the time of accident.

[0237] Mark as normal or abnormal: If the vehicle driving trajectory conforms to the normal driving behavior pattern, mark this driving trajectory data as normal. If there are abnormal situations in the vehicle driving trajectory, mark this driving trajectory data as abnormal.

[0238] Example illustration:

[0239] Vehicle insurance policy information:

[0240] Vehicle identification number: LJWFF3A1XK1234567

[0241] License plate number: Shanghai A12345

[0242] Insured: Li Si

[0243] Policy amount: 20,000 yuan

[0244] Insured date: 2022-01-01

[0245] Date of accident: 2022-08-15

[0246] Indemnity amount: 15,000 yuan.

[0247] Driving trajectory data:

[0248] Driving speed:

[0249] Time point 1: 40 km / h

[0250] Time point 2: 100 km / h

[0251] Time point 3: 80 km / h

[0252] Driving route:

[0253] Starting point: Huangpu District, Shanghai

[0254] End point: Minhang District, Shanghai

[0255] There were abnormal detours and deviations from the route along the way.

[0256] Duration:

[0257] Stayed in a remote area for 2 hours.

[0258] Driving behavior:

[0259] There were sudden braking and sharp turns before the accident.

[0260] Process of marking anomalies:

[0261] Determine the insurance policy information corresponding to the vehicle driving trajectory data: match the vehicle driving trajectory data with the corresponding auto insurance policy information according to the vehicle frame number LJWFF3A1XK1234567.

[0262] Determine whether the vehicle driving trajectory data is normal:

[0263] Driving speed: On the date of the accident, 2022-08-15, the vehicle's speed record showed one speeding behavior (100 km / h), which exceeded the speed limit.

[0264] Driving route: According to GPS data, the vehicle took abnormal detours and deviated from the route on the day of the accident, which did not conform to the normal driving route.

[0265] Duration of stay: The vehicle stayed in a remote area for 2 hours, which was too long and in an unusual location.

[0266] Driving behavior: There were sudden braking and sharp turns before the accident, which are usually related to emergency situations or abnormal driving.

[0267] Marking exceptions:

[0268] Based on the above analysis, it is determined that there is an abnormality in the vehicle driving trajectory, and the driving trajectory data is marked as abnormal.

[0269] Through the above steps, this embodiment can improve the transparency and accuracy of the auto insurance claims process and prevent abnormal and fraudulent behaviors.

[0270] The present invention also provides a device for constructing a vehicle insurance anomaly detection model.

[0271] Reference Figure 3 , Figure 3 This is a functional module diagram of the first embodiment of the construction device of the vehicle insurance anomaly detection model of the present invention. The construction device of the vehicle insurance anomaly detection model includes:

[0272] An information acquisition module, which acquires the automobile insurance policy information, acquires the vehicle driving trajectory data corresponding to the automobile insurance policy information, and marks the vehicle driving trajectory data as normal or abnormal;

[0273] A feature extraction module, extracting claim settlement features of the motor vehicle insurance policy information, and extracting trajectory features according to normal or abnormal labels of the vehicle driving trajectory data;

[0274] The model building module uses the isolation forest algorithm to build a car insurance anomaly detection model according to the claim characteristics and the trajectory characteristics.

[0275] In addition, an embodiment of the present invention further provides a storage medium.

[0276] The storage medium stores a program for constructing a vehicle insurance anomaly detection model. When the program for constructing a vehicle insurance anomaly detection model is executed by a processor, the steps of the method for constructing a vehicle insurance anomaly detection model as described above are implemented.

[0277] The specific implementation of the storage medium of the present invention is basically the same as the various embodiments of the method for constructing the above-mentioned vehicle insurance anomaly detection model, and will not be repeated here.

[0278] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the enlightenment of the present invention, ordinary technicians in this field can also make many forms without departing from the scope of protection of the present invention and the claims. All equivalent structures or equivalent process changes made using the contents of the specification and drawings of the present invention, or directly or indirectly used in other related technical fields, are protected by the present invention.

[0279] In the embodiments provided in the present application, it should be understood that the disclosed devices and methods can also be implemented in other ways. The device embodiments described above are merely schematic. For example, the flowcharts and block diagrams in the accompanying drawings show the possible architecture, functions and operations of the devices, methods and computer program products according to the multiple embodiments of the present application. In this regard, each box in the flowchart or block diagram can represent a module, a program segment or a part of a code, and the module, a program segment or a part of a code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or the flowchart, and the combination of boxes in the block diagram and / or the flowchart can be implemented with a dedicated hardware-based system that performs a specified function or action, or can be implemented with a combination of dedicated hardware and computer instructions.

[0280] In addition, each functional module in each embodiment of the present application can be integrated together to form an independent part, or each module can exist separately, or two or more modules can be integrated to form an independent part. If the function is implemented in the form of a software functional module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or the part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, RandomAccess Memory), a disk or an optical disk.

[0281] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0282] The above are only various implementations of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A method for constructing a car insurance anomaly detection model, characterized in that: The method for constructing the vehicle insurance anomaly detection model comprises the following steps: Obtaining automobile insurance policy information, obtaining vehicle driving trajectory data corresponding to the automobile insurance policy information, and marking the vehicle driving trajectory data as normal or abnormal; extracting the claim settlement features of the motor vehicle insurance policy information, and extracting the trajectory features according to the normal or abnormal labels of the vehicle driving trajectory data; An auto insurance anomaly detection model is constructed using the isolation forest algorithm according to the claim characteristics and the trajectory characteristics.

2. The method for constructing a vehicle insurance anomaly detection model according to claim 1, characterized in that: The step of constructing a car insurance anomaly detection model using an isolation forest algorithm according to the claim characteristics and the trajectory characteristics comprises: Using the claim settlement feature and the trajectory feature as a data point set of an isolation forest model, and initializing the number of isolated trees to a preset value; For each isolated tree, randomly sampling a data point from the data point set for constructing the isolated tree; For each data point, calculate its isolation score on each isolated tree; The average score of all data points is calculated based on the isolation score of each data point. If the difference between the isolation score of a data point and the average score is within a preset difference anomaly threshold, the data point is marked as abnormal, and the auto insurance anomaly detection model is constructed.

3. The method for constructing a vehicle insurance anomaly detection model as claimed in claim 2, characterized in that: Before the step of marking a data point as abnormal if the difference between the isolation score of a data point and the average score is within a preset difference abnormality threshold, the step further includes: Set the preset difference abnormality threshold to a range of 0-15, or set the preset difference abnormality threshold to a range of greater than 100.

4. The method for constructing a vehicle insurance anomaly detection model according to claim 2, characterized in that: Before the step of building the vehicle insurance anomaly detection model, the method further includes: The isolation forest model is trained cyclically until the number of training times for each data point in the data point set reaches a preset number of times.

5. The method for constructing a vehicle insurance anomaly detection model according to claim 1, characterized in that: After the step of constructing a car insurance anomaly detection model using an isolation forest algorithm according to the claim settlement features and the trajectory features, the method further includes: A test sample set is obtained, wherein the ratio of positive and negative samples in the test sample set is determined according to the motor vehicle insurance policy information, and the performance of the motor vehicle insurance anomaly detection model is evaluated using the test sample set.

6. The method for constructing a vehicle insurance anomaly detection model according to claim 1, characterized in that: The steps of obtaining the automobile insurance policy information include: The motor vehicle insurance policy data of a preset year is obtained, the motor vehicle insurance policy data is cleaned, duplicate data and data with missing information in the motor vehicle insurance policy data are deleted, and the motor vehicle insurance policy information is obtained.

7. The method for constructing a vehicle insurance anomaly detection model according to claim 1, characterized in that: The step of marking the vehicle driving trajectory data as normal or abnormal comprises: Determine the insurance policy information corresponding to the vehicle driving trajectory data, and judge whether the vehicle driving trajectory corresponding to the insurance policy information is normal when the insurance policy information occurs; If so, marking the vehicle driving trajectory data as normal; If not, the vehicle driving trajectory data is marked as abnormal.

8. A device for constructing a vehicle insurance anomaly detection model, characterized in that: The construction device of the vehicle insurance anomaly detection model includes: An information acquisition module, which acquires the automobile insurance policy information, acquires the vehicle driving trajectory data corresponding to the automobile insurance policy information, and marks the vehicle driving trajectory data as normal or abnormal; A feature extraction module, extracting claim settlement features of the motor vehicle insurance policy information, and extracting trajectory features according to normal or abnormal labels of the vehicle driving trajectory data; The model building module uses the isolation forest algorithm to build a car insurance anomaly detection model according to the claim characteristics and the trajectory characteristics.

9. A device for constructing a vehicle insurance anomaly detection model, characterized in that: The construction device of the vehicle insurance anomaly detection model includes a memory, a processor, and a construction program of the vehicle insurance anomaly detection model stored in the memory and executable on the processor. When the construction program of the vehicle insurance anomaly detection model is executed by the processor, the steps of the construction method of the vehicle insurance anomaly detection model as described in any one of claims 1 to 7 are implemented.

10. A storage medium, characterized in that: The storage medium stores a program for constructing a vehicle insurance anomaly detection model. When the program for constructing a vehicle insurance anomaly detection model is executed by a processor, the steps of the method for constructing a vehicle insurance anomaly detection model as described in any one of claims 1 to 7 are implemented.

Images